Vulnerability of a Top Marine Predator in One of the World’s Most Impacted Marine Environments (Arabian Gulf)

Bruno Diaz Lopez (  bruno@thebdri.com ) Bottlenose Dolphin Research Institute BDRI https://orcid.org/0000-0002-0388-3289 Séverine Methion Bottlenose Dolphin Research Institute BDRI Himansu Das EAD: Environment Agency Abu Dhabi Ibrahim Bugla EAD: Environment Agency Abu Dhabi Maitha Al Hameli EAD: Environment Agency Abu Dhabi Hind Al Ameri EAD: Environment Agency Abu Dhabi Ahmed Al Hashmi EAD: Environment Agency Abu Dhabi Edwin Grandcourt EAD: Environment Agency Abu Dhabi


Introduction
As relatively large top predators, dolphins are key component of marine ecosystems. These small cetaceans ful l ''umbrella'' and '' agship'' criterion and are of high ecological value (Connor et al. 2000). Bottlenose dolphins (Tursiops spp.), despite being widely considered some of the most adaptable of the world's cetaceans, are highly susceptible to environmental changes (Bejder et al. 2006;Sprogis et al. 2018;Díaz López 2019;Methion and Díaz López 2019). Due to their inshore distribution and life history characteristics (i.e., relatively large size, slow growth, late maturation, long gestation period, single births at a time, and long calving interval), bottlenose dolphins are vulnerable to a range of anthropogenic impacts such as habitat modi cation, over shing, noise and chemical pollution, bycatch, and boat strikes (Díaz López 2006;Wang and Yang 2009).
The International Union for the Conservation of Nature (IUCN) has recently assigned the Indo-Paci c bottlenose dolphin (Tursiops aduncus, hereafter bottlenose dolphin) the category of "Near Threatened" in light of its nearshore distribution, local declines, and of the increasing intensity of threats to the species (Braulik et al. 2019). There is however a lack of information on habitat use and population abundance throughout much of the species' range, and it is likely that some bottlenose dolphin populations could be classi ed as "Threatened", particularly in habitats where rapid recent economic, social, and industrial development has not been adequately compensated by conservation measures (Braulik et al. 2019).
The Arabian Gulf (also known as the Persian Gulf and referred to hereafter as the Gulf) represents one of the most extreme and anthropogenically impacted marine environments (Halpern et al. 2008). The Gulf supports a wide variety of marine ecosystems (including seagrass beds, mangroves, coral reefs, and marshes) that are uniquely adapted to extremes of sea surface salinity and temperature and low levels of primary production (Sheppard et al. 2019). Coastal environments in the Gulf are affected by intensive dredging and reclamation activities and various sources of noise and chemical pollution (e.g. seismic surveys for oil and gas, marine tra c, industrial waste, brackish wastewater, ports and re neries, oil spills and domestic wastewater) (Sheppard et al. 2010;Vaughan et al. 2019). Although it is well known that the extreme environmental conditions and anthropogenic activities affect the diversity, abundance, and distribution of many marine species in this region (Sheppard et al. 2010), there is a lack of information on top marine predators' ecology. In particular, there is a paucity of data on bottlenose dolphin ecology, a species regularly present in the Gulf, which prevents a comprehensive assessment of its conservation status (Baldwin et al. 1999;Preen 2004;. Obtaining further information on bottlenose dolphin ecology such as on habitat use, site delity, abundance estimation, and potential overlapping with human activities would therefore be essential to ensure the persistence of this species in these waters. With little information available on the distribution and abundance of cetaceans in the Gulf, a collaborative project between the Environment Agency -Abu Dhabi (EAD) and the Bottlenose Dolphin Research Institute (BDRI) was initiated in 2014 to study cetacean ecology in the Southern Arabian Gulf (Abu Dhabi Emirate, United Arab Emirates UAE). In this study, the objective was to provide novel information on bottlenose dolphin ecology and vulnerability to human activities in Abu Dhabi waters. In particular, we identi ed important habitats for bottlenose dolphins and the environmental factors that affect their presence and abundance in the area. This was achieved by examining the relationships between environmental variables and bottlenose dolphin presence and relative abundance. Further, we assessed for the rst time the abundance of this species in the region and we identi ed individual patterns of movement and site delity through the use of mark-recapture methods. Based on the ndings, we made recommendations to support bottlenose dolphins' conservation in the Southern Arabian Gulf.

Study area
This study was conducted along Abu Dhabi Emirate's shoreline. This coast represents about 76 percent of the UAE's Arabian Gulf coast (Abdessalaam 2007). The study area, located between 24.808° N 51.840° E and 24.857° N 54.858° E, comprises approximately 25,000 km 2 (Fig. 1). Due to the high geographical latitude, relatively shallow depth and high evaporation rates, the study area is characterized by extreme environmental conditions. This area is in uenced by atmospheric processes associated with the winds known as Shamal (i.e. strong, dry and cold northwesterly winds). Shamal winds are stronger during the winter months than during the summer months, being responsible for a drastic reduction in sea surface temperature (oscillating from 15°C in winter to 36°C in summer) (John et al. 1990). These coastal waters are also characterized by extreme salinity values that can exceed 48 psu (Vaughan et al. 2019). This area has a slightly sloping shelf and comprises different habitats: an extensive coastal sabkha (supra-tidal saline zone, Evans et al. 1969), marshes, mangroves, seagrasses, coral reefs, offshore and coastal islands, the latter forming channel systems. Mangroves, algae and seagrasses cover extensive areas that provide shelter and forage for a multitude of marine species (Abdessalaam 2007). Three cetacean species are present throughout the year: bottlenose dolphins, Indian Ocean humpback dolphins (Sousa plumbea) and nless porpoises (Neophocaena phocaenoides) . While 13.45% of Abu Dhabi's marine area is currently protected, these coastal waters are experiencing high anthropogenic impact, mainly in the form of the oil and energy industry, and the expansion of the human population leading to land reclamation, port construction, and increased boat tra c and shing effort (Sheppard et al. 2010;Al Dhaheri et al. 2017).

Data collection
The study area was divided into three different sub-sections based on ecological characteristics and according to logistical constraints (e.g. accessibility, ship launching facilities) (following Díaz López et al. 2018): 1. Eastern Region (between 24 808′N 54 800′E and 24 857′N 54 858′E, about 4 000 km 2 ). This region faces the greatest anthropogenic pressure and includes the city of Abu Dhabi (about 1.4 million inhabitants) and inshore islands that delimit a series of natural and man-made channels, seagrass beds, mangrove areas and, in the open coastal waters, coral reefs. This region includes three small Marine Protected Areas: Al Saadiyat (59 km 2 ), Bul Syayeef (145 km 2 ), and Ras Ghanada (54 km 2 ).
3. Western region (between 24,808°N 51,840°E and 24,857°N 53,800°E, 11 000 km 2 ). This region is the one with the lowest human population density and the highest diversity of habitats (including extensive seagrass beds, coastal and offshore islands, and coral reefs). It includes the Al Yasat marine protected area (2 083 km²), The surveys were conducted using a 45-foot custom research vessel powered by two 300-hp outboard engines. We established a systematic daily survey route with transects adapted to the speci c conditions of each region, taking into consideration that the vessel was departing from a different harbour within each region. It was not possible to follow a zigzag pattern because the channels, islands, and shallow waters conditioned the trajectory of the transect lines. Each region was monitored for a minimum of three daily surveys for each of the six sampling seasons. A sampling season lasted between 15 and 21 days.
The spatial distribution of the effort varied according to weather conditions and time constraints throughout the study period.
To reduce bias in our ability to detect dolphins, surveys were conducted when visibility was not reduced by rain or fog, wind strength was < 3 on the Beaufort scale, and wave height was < 0.2 m. Surveys were carried out during daylight hours at a constant speed (between 8 and 10 knots). At least three experienced observers, located on an upper observation deck (2.5 meters above sea level), were conducting 360 degrees scan in search of dolphins (with the naked eye and/or with 10 × 40 binoculars). In order to include seasonality as a factor in the analysis, sampling seasons were conducted in different months of the year.
On each daily boat-based survey, the date, time, GPS position, boat speed, and environmental data were recorded as an instantaneous sample every 20 minute (Díaz López and Methion 2018). The spatial resolution of this 20-minute interval was 3 nm (given a speed of 8-10 knots) and dolphin detection/no visual detection was recorded instantly for all 20-minute sampling points. When dolphins were sighted, searching effort ceased and the vessel slowly maneuvered toward them to minimize possible disturbance during approach. A group of bottlenose dolphins was de ned as one or more individuals observed within a 100-m radius and, if more than one, interacting with each other and performing the same behavioural activity . At least two observers with digital SLR cameras equipped with telephoto lenses attempted to photograph both sides of the dorsal n of each dolphin of the group. Group size and composition were estimated based on the total count of individuals observed at a given time in the area, and the data were later veri ed during the analysis of the photographs . The age of the individuals was determined as dependent calves or adults, based on behavioural cues and visual size assessment (Díaz López and Methion 2017). At the end of an encounter, the searching effort continued along the previously planned route.

Environmental predictors
Twelve environmental predictors were considered for each 20-minute sample: date (year and month), time (UTC in hours), latitude and longitude coordinates, depth (m), distance to coast (m), sea surface temperature (SST in ° Celsius), sea surface salinity (SST in psu), chlorophyll-a concentration (in mg×m − 3 ) during the day of sampling, chlorophyll-a concentration one month before the day of sampling, and marine benthic habitat type. The depth was extracted from a 30 arc second bathymetric map of the General Bathymetric Chart of the Oceans (GEBCO Compilation Group 2020). The minimum distance of each 20-minute sample from the coast was calculated with the NNJoin plug-in in QGIS 2.18. SST and chlorophyll-a data were obtained as 8-day rasters, with a spatial resolution of 4 km × 4 km from the Giovanni online data system (Acker and Leptoukh 2007). The SSS was obtained as monthly rasters, with a spatial resolution of 0.5 degrees, from the COPERNICUS Marine Environment Monitoring Service website (http://marine.copernicus.eu). The marine habitat type was obtained from a 5-m resolution Quick Eye Image created by the Environment Agency -Abu Dhabi (EAD), including six different types: coral reef, deep subtidal sea oor, dredged areas, hard bottom, seagrass bed, and unconsolidated bottom. The "point sampling" tool in QGIS 2.18 was then used to extract raster values from multiple layers in each 20-minute sample in order to link bottlenose dolphin relative presence or abundance to environmental predictors.

Data analysis and modelling framework
For the spatial analysis of the observation effort, the study area was divided into 3 nm hexagonal cells creating a polygon grid using the QGIS software. The size and shape of the cells were designed to t both the visual area of the research vessel and the distance covered between each 20-min sample ). The number of 20-min samples collected within a cell was considered a fair representation of the survey effort.
A generalized additive modelling (GAM) framework was used to explore the predictors that could have affected the two response variables selected in this study: the presence (habitat use) and number of bottlenose dolphins (relative abundance). GAMs are widely used to interpret ecological interactions and are particularly suited to the type of non-linear responses expected in speciesenvironment relationships (Hastie and Tibshirani 1990). The data exploration protocols described by Zuur et al. (2010) were used to identify outliers, data variability, and relationships between predictors and the response variable. Modelling was initiated using a General Linear Model (GLM), which included 12 covariates (latitude and longitude, year, month, observation effort, depth, distance from shore, SST, SST, chlorophyll-a concentrations during the day of sampling and one month prior to sampling, and marine benthic habitat type) that could potentially drive the response variable. Before tting the model, possible co-linearity between the predictors was investigated by calculating the Spearman correlation coe cients in pairs (r) and the variance in ation factors (VIF). When the variables showed a high correlation (above r = 0.7 and VIF > 3) they were not used together in the same model (Dormann et al. 2013).
To nd a set of explanatory variables that do not contain collinearity, the variables were eliminated one by one and the VIF values were then recalculated. Following this procedure, the month of the year was excluded before starting the model t as it was related to other variables which were instead included due to their biological interpretability (SST and SSS, Kruskall-Wallis test, P < 0.01).
The use of two types of GAMs in this study, with presence-absence data and relative abundance data, allowed an accurate prediction of the response variables (Howard et al. 2014). To choose the most appropriate presence-absence model to address an apparently zero-in ated dataset, three different models such as GAMs with logistic link function, Tweedie or Negative Binomial distributions were compared using the Akaike Information Criterion (AIC) ). The number of bottlenose dolphins was modelled using a GAM with a negative binomial distribution and logarithmic link function. The smooth functions were constructed as cubic splines and their optimal shape were estimated by minimising the general cross validation (GCV) criterion. The optimal model was selected using a combination of backward and forward model selection procedures based on the corrected Akaike Information Criteria (AICc). Model assumptions were checked by visual inspection of the residuals and regression ts were examined using plots of residuals against tted values. The nal model was the model with the lowest AICc given that effects of all explanatory variables retained in the model were statistically signi cant and there were no clear patterns in the residuals (Hastie and Tibshirani 1990). The GAMs results and diagnostic information about the tting procedure were implemented from the mgcv (Wood, 2006) and MASS (Venables and Ripley 2002) packages in v. 1.8.1. of the statistics and graphics tool R (R Development Core Team, 2011). The Durbin-Watson test (from the R package "lmtest", Zeileis and Hothorn 2002) and auto-correlation functions (ACF) were used to check for serial correlation, both in the raw data and in the residuals from the models. Partial predictions with 95 % con dence intervals were plotted for each covariate included within the nal model. The data are presented as means ± standard error (SE). To determine the areas of highest predicted probability for the presence of bottlenose dolphins, partial predictions in R were calculated using the nal model. For variables that vary over time (e.g. SST, SSS), mean values were calculated. Predicted values were made on the response scale (between 0 and 1) and displayed on a map using the centroids of 3 nm x 3 nm hexagonal cells.

Analysis of photographs and mark-recapture abundance models
Bottlenose dolphins were identi ed based on the natural markings present on their dorsal ns following the methods of selection and photo-identi cation analysis described in Methion and Díaz López (2018). Only photographs with good and excellent quality conditions were used. Likewise, only distinctly marked adult bottlenose dolphins were included in the photo-identi cation analysis.
Photographs containing calves (immature and newborn dolphins) and unmarked individuals were excluded. Using POPAN in SOCPROG 2.8, abundance estimates were calculated and tted to four open population mark-recapture models described in Whitehead (2009) The selection of the most appropriate model was based on the lowest AIC.
Total abundance was calculated using estimates generated from the most appropriate model and corrected for the proportion of distinctly marked individuals in the population. To calculate the proportion of distinctly marked individuals in the population, the number of distinctly marked adult individuals was divided by the total number of individuals observed in each group, averaged across all encounters . The 95% con dence intervals were calculated using the "delta method" (Seber 1982).

Survey effort and presence of bottlenose dolphins
Between June 2014 and November 2019, 80 daily surveys were conducted covering 9 933 km (Table 1). In total, 527 hours were spent in satisfactory conditions and 1 547 samples were collected. Overall, 89 groups of bottlenose dolphins were encountered (Fig. 1) Mann-Whitney, P < 0.001). In two encounters throughout the study, bottlenose dolphins were observed in mixed feeding aggregations with Indian Ocean humpback dolphins.

Environmental drivers of bottlenose dolphin presence
The most parsimonious model to t the data was a GAM with a logistic link function with eight variables (Table 3). The GAM explained 22.6% of the variation in bottlenose dolphin presence (AICc 31.1 units lower than the initial model). Bottlenose dolphin presence was predicted to be signi cantly in uenced by location (latitude, longitude, and distance from shore), SST, and chlorophyll-a concentration one month prior to the sampling date (Supplementary Information S1). It was predicted that bottlenose dolphin occurrence was more likely in the central and western regions of Abu Dhabi (Fig. 2). Observation effort, SSS, and chlorophyll-a concentration during the sampling date did not signi cantly contribute to the observed variation in bottlenose dolphin occurrence (P > 0.05).

Environmental drivers of bottlenose dolphin relative abundance
Based on AICc scores, the most parsimonious model to t the data was a GAM with a negative binomial distribution with ve variables (Table 4). The GAM explained 47 % of the variation in bottlenose dolphin relative abundance (AICc 161 units lower than the initial model). Bottlenose dolphin relative abundance was predicted to be signi cantly in uenced by the location (latitude, longitude, and type of marine benthic habitat), SSS, and chlorophyll-a concentration one month prior to the sampling date (Supplementary Information S2). Bottlenose dolphin relative abundance was predicted to be higher in the western and central regions of Abu Dhabi.
The observed aggregations of bottlenose dolphins were smaller in locations with seagrass bed and hard bottom than those aggregations observed in other types of marine benthic habitat (Fig. 3).  The most appropriate open model showed a population that was declining at a constant rate (Table 5). Population size, mortality rate, and population decline per sampling period were estimated using maximum likelihood. Abundance estimates suggested a population size of 563 ± 117 (95% CI = 358-932) distinctly marked adult individuals. Based on the proportion of distinctly marked adult bottlenose dolphins (72%), 782 (95% CI = 497-1 294) bottlenose dolphins were estimated to inhabit Abu Dhabi waters. Notations: n = estimated population size; CI = con dence interval; * Bootstrapped (n = 500) SOCPROG Model results; n.a = not available; m = estimated mortality rate; T = estimated trend; e = estimated emigration rate, r = estimated reimmigration rate, Nc = number of animals captured; s.e. = standard error; s.p.= number of sampling periods; Ø = ratio of marked adult bottlenose dolphins; Nt = estimate of total population size after correcting for proportion of identi able individuals; AIC = Akaike Information Criterion.

General importance and impact
Studies that assess marine top predator habitat use and abundance are fundamental in areas such as the Gulf, where many species live near to their tolerances and are highly impacted by human activities (Sheppard et al. 2010). Our study provides novel information about bottlenose dolphin habitat use, movements, and abundance along Abu Dhabi Emirate coastline. The value of this type of study is well-recognised to identify areas of biological importance and to determine the spatio-temporal scale at which human activities may impact on dolphins, therefore facilitating their conservation (Cheney et al. 2014).
On a broader scale, this study con rms the strong adaptability and tolerance of bottlenose dolphins to extreme environmental conditions within a highly heterogeneous and impacted marine habitat. From a regional perspective, we provide the rst estimates of reef sh, as well as cephalopods (Amir et al. 2005) and SST, SST, and zooplankton abundance substantially affect the availability of these species (Houde et al. 1986;Grandcourt et al. 2005). The relationship between a relatively low concentration of chlorophyll-a concentration one month before the day of sampling and bottlenose dolphin presence could be interpreted as a period of phytoplankton decline mediated by zooplankton grazing and, consequently, the increase of zooplanktivorous sh species. Along Abu Dhabi coast, pelagic sh species are most abundant between September and May, and sh generally move in an east-west direction during this period (Grandcourt et al. 2005). Reef sh communities also show seasonal uctuations with an abundance peak when water temperature is colder (Grandcourt et al. 2011;Grandcourt 2012). Likewise, seaweed beds are more abundant and mix with sea grasses between September and May, which provide a vital settlement and breeding ground for various sh species (George and John 1999).
Our study cannot conclusively prove the direct impact of human disturbance on bottlenose dolphins due to the in uence of other explanatory factors such as changes in prey availability. The positive relationship between bottlenose dolphin habitat use and relative abundance and distance from the coast (human settlements) could however be related to dolphin avoidance of elevated anthropogenic activities in coastal waters. Likewise, the higher preferences of bottlenose dolphins for the central and western regions are likely due to less human disturbance in these areas compared to the eastern region (less noise pollution; less marine tra c; less habitat degradation). Compared to the western and central regions, the eastern coast of Abu Dhabi has a higher human population density, has experienced a more rapid population increase, and has fewer marine protected areas. In this region, coastal dredging and development for industrial, commercial, and residential use have induced dramatic changes in marine ecosystems which may indirectly decrease the availability of dolphin's prey species (Sheppard et al. 1992(Sheppard et al. , 2010Al Dhaheri et al. 2017). Other threats such as pollution and noise are also more prevalent in eastern coastal waters, some of which are di cult to quantify alone or in synergy with others (Gordon et al. 2003). This suggests a potentially signi cant impact of anthropogenic disturbance on bottlenose dolphins.

Abundance estimation
Our commercial shing is prohibited, except by artisanal shermen using traditional gear, and the capture of any dugong, turtle or marine mammal is prohibited (UAE Federal Law No. 23/1999). Dredging, land lling or other development activities on the coast are also restricted. However, the small size of these marine protected areas, taking into account the distribution and observed movements of bottlenose dolphins, is a clear limitation to contribute to the conservation of suitable habitats for the species. The observed movement of several individual bottlenose dolphins between sites separated by > 250 km of coastline suggests that movements along these distances are not uncommon for this species. In addition, the lack of information on cetacean bycatch in Abu Dhabi waters makes it di cult to assess the magnitude and population-level impact of sheries in the area. Future research efforts should focus on assessing the effects of artisanal shing gear, particularly traps and gillnets, on bottlenose dolphin populations.
In addition to preserving and increasing marine protected areas within these waters and conducting future studies on the impact of sheries on dolphin populations, regional connectivity should be of particular value for bottlenose dolphins and other coastal cetacean species since alteration of their coastal habitats can result in population declines and eventual local or regional extinctions (Barlow et al. 2010;Mei et al. 2012). Our results shed light on the importance of transnational research on the distribution of cetaceans in the waters of the Gulf for the establishment of trans-frontier conservation areas (TFCAs). TFCA countries under the umbrella of the Gulf Cooperation Council Biodiversity Committee (GCCBC) should develop and implement regional action and management plans for the conservation of marine mammal species across borders. The oceanography of the Gulf and projections of the future climate of the region are su cient reasons to believe that coastal cetacean species are seriously threatened by anthropogenic activities. We therefore recommend further research to identify important corridors for cetaceans within the coastal waters of the Gulf and establish collaboration between researchers and different stakeholders to ensure their integration into management plans.

Declarations
Funding: This research has been funded by the Environment Agency -Abu Dhabi (EAD).
Con ict of interest/Competing interests: The authors declare that they have no known competing nancial interests or personal relationships that could have appeared to in uence the work reported in this paper.
Availability of data and material: Data will be provided under request.
Code availability: R Script will be provided under request. Consent to participate: All authors gave nal approval to participate.
Consent for publication: All authors gave nal approval for publication.