Population ecology and the management of whale watching operations on a data‐deficient dolphin population

Abstract Whale watching is a popular commercial activity, producing socio‐ecological benefits but also potential long‐term effects on the targeted cetacean population. This industry is currently developing in data‐deficient contexts in a largely unregulated fashion. Management schemes should adopt precaution and be informed by the relevant literature, but would be more effective if the assessment of the target population vulnerability, biological impacts, and management implications was drawn from site‐specific data. This paper focuses on a reef‐associated, data‐deficient population of spinner dolphins in the Egyptian Red Sea. In Satayah Reef, new information on population size and dynamic parameters were documented using visual observation and photo‐identification‐based capture–recapture methods (Cormack–Jolly–Seber time‐since‐marking model). Dolphins occurred on 98% of the survey days. Average school size was 66 individuals (±42.1 SE), with most groups including calves. The population was equally divided into recurrent and transient individuals. An “emigration + mortality” model best described residence at the site. Five recurrent males (5% of the Satayah population) provided connectivity between this and the geographically close population of Samadai Reef. Average annual survival probability was 0.83 (±0.06 SE) in the year following first capture and 0.99 (±0.06 SE) for recurrent individuals. Mean yearly population sizes ranged 143–207 individuals. The study had the power to detect a 30% decline in the population, but not the rate of change in abundance estimated from the data (r = 0.018 ± 0.04), which would have required a 3‐ to 5‐times longer study. Synthesis and application: These findings advance the assessment of the Satayah population's intrinsic vulnerability and have three major management applications: (a) the delineation of management units; (b) the identification of key indicators for future impact monitoring and assessment; and (c) realistic estimates of the statistical power for trend detection. Based on our results, we recommend supporting future research, devising site‐specific time–area closure plans, and integrating them in a regional scheme. Approaches employed in this case study can inform the management of whale watching industries targeting other data‐deficient populations.

In order to manage WW operations effectively, site-specific information on the vulnerability of individuals and populations targeted are needed (Higham, Bejder, & Lusseau, 2009). The importance of such information is trifold. In a developing WW scenario, site-specific information allows the assessment of the vulnerability of the targeted cetacean population or subpopulation. Indicators of vulnerability include biological and ecological conditions that regulate individual exposure, sensitivity, and recovery to human interactions (De Lange, Sala, Vighi, & Faber, 2010;De Lange, Van der Pol, Lahr, & Faber, 2005), including age, sex and reproductive classes, body condition, behavior, frequency of exposure to interactions (Christiansen & Lusseau, 2014), and other species-or site-specific features. A combination of observational and photo-identificationbased capture-recapture (CR) studies can provide such information (Cribb, Miller, & Seuront, 2012;Karczmarski et al., 2005;Norris et al., 1994;Parra, Corkeron, & Marsh, 2006). Individual cetaceans are often recognized from the marks that naturally accumulate on or near the dorsal fin, and their occurrence in the study area is recorded by means of photo-identification (photoID), a commonly used technique to collect photographic evidence of the individuals encountered (Hammond, Mizroch, & Donovan, 1990). The capture histories of distinctive individuals (i.e., vectors of their presence and absence at sampling occasions) are analyzed in CR models to estimate individual site fidelity and population parameters (Hammond, Mizroch, & Donovan, 1990;Kendall, Pollock, & Brownie, 1995;Otis et al., 1978;Pollock, 2000;Seber, 1982). Among these parameters, residence, female reproductive rate, individual survival, and population size have been proposed as valid metrics to assess the biological impacts of WW activities (Bejder et al., 2006;Lusseau, Slooten, & Currey, 2006). Population ecology can also help monitor the efficacy of implemented measures in safeguarding wild populations (Gormley et al., 2012). Finally, site-specific studies can support management and decision-making processes through the identification of targets of protection (De Lange et al., 2010), diagnostic indicators for adaptive management (e.g., Limits of Acceptable Change (LAC); Stankey et al., 1985;Duffus & Dearden, 1990;Higham, Bejder, & Lusseau, 2009), supporting future research, devising site-specific time-area closure plans, and integrating them in a regional scheme. Approaches employed in this case study can inform the management of whale watching industries targeting other data-deficient populations.

K E Y W O R D S
CJS models, inequality model, lagged identification rate, Red Sea, spinner dolphin, tourism management, whale watching F I G U R E 1 A group of spinner dolphins in a resting area off the Egyptian coast (Photo by A.Cesario/HEPCA) and considerations on effective study designs (e.g., Gerrodette's inequality model;Gerrodette, 1987).
The management plan in Samadai Reef substantially reduces behavioral disruptions caused by human interactions, which are instead documented as pervasive and severe at the nonmanaged resting area at Satayah Reef (Fumagalli et al., 2018). There, in 2009, tourism was reported as "opportunistic" with a potential for further development (O'Connor et al., 2009). Indeed, as of 2014, 90+ swimmers and 10 inflatable boats could simultaneously approach a resting school during dedicated swim-with activities, and the active, invasive interactions could last for up to 9 hr daily (Fumagalli, 2016 Visual observations to detect the presence of dolphins were carried out from a dedicated, stationary vessels moored in the western lagoon of Satayah Reef. Observations started at dawn, or at arrival on site, and ended at sunset, or when research efforts were interrupted for logistical reasons. The sighting of the first dolphin in the lagoon from the stationary vessel marked the beginning of an "encounter," which ended with the departure of the last dolphin, or with the end of the daily observations. Given the structure of the reef protecting the lagoon from the mainly northerly winds, all surveys were conducted in calm sea conditions (Beaufort sea state <2) even on high wind days.

| School composition
School size (total number of individuals in the lagoon) and composition in age classes were estimated in 2010-2013 during 35 photographic sessions (see below). An individual was considered "calf" if <¾ the size of an adult and in regular association with an adult, or "newborn calf" if it showed obvious fetal folds (Norris et al., 1994).
All other individuals were "adult." The occurrence and number of females in early or advanced pregnancy stage (see in Appendix 1) were assessed visually during underwater sessions in 2011-2013. Independent field estimates provided by experienced researchers were averaged to estimate the mean school size, and the number of calves and newborn calves in a school.

| Photographic identification
When dolphins were detected in the lagoon of Satayah Reef, a first photographic session was carried out for photoID purposes. If the photographers deemed the coverage of the school insufficient, or when the encounter extended throughout the day, at least one more photographic session was performed (ideally one in the morning and one in the afternoon) in order to increase opportunities to cover the entire school and to account for possible changes in the daily school composition. The duration and number of sessions were also context-dependent: when the co-occurrence of tourism activities made it difficult to maneuver around the dolphin school, causing concerns over the quality of data collected as well as the welfare of the animals, sessions were interrupted and resumed at a later stage.
Photographic sessions did not follow pre-established line transect and were aimed to provide even coverage of all individuals in each group found in the lagoon. Sessions were conducted from the surface, on board 4-to 6-m inflatable boats equipped with 45-150 HP outboard engines, and/or underwater, snorkeling in proximity to the dolphins. Underwater photoID was shown to provide good coverage of the dolphin group (0.84 ± 0.15 SD; Cesario, 2017) and deliver information not available from the surface, thus was preferred over boat-based photoID when conditions allowed. In both cases, photographers attempted to equally sample all individuals and groups in the lagoon, irrespective of their distinctiveness, behavior, sex and age, and followed a code of conduct to minimize disturbance to the school (details in Appendix 1).
All sessions carried out in the same day, hence on the same encounter, were pooled together in a photographic occasion. To promote consistent and higher quality assessment of individual presence (see Ottensmeyer & Whitehead, 2003), only 30 occasions with a number of photographs at least three times the estimated school size were retained for further analyses (7 in 2010, 7 in 2011, 7 in 2012, and 9 in 2013). In addition, four occasions from 2006 were included in the creation of the catalogue of individuals and in the assessment of individual site fidelity to provide historical perspective. PhotoID images were assessed by experienced researchers for photographic quality and individual distinctiveness using protocols modified from the literature (Friday, Smith, Stevick, & Allen, 2000;Urian et al., 2015) and consistent with studies on the Samadai population (Cesario, 2017; see Appendix 1). Very Distinctive (D1), TA B L E 1 Summary of the sampling effort per year: survey dates (Jun = June, Jul = July, Aug = August), sampling effort, number of encounters, number of photoID sessions matching the criterion for inclusion in the analyses (highlighted in bold in "Survey dates") and number of distinctive individuals identified each year

| Site fidelity
The lagged identification rate (LIR) was estimated in SOCPROG 2.7 (Whitehead, 2015) to test scenarios in which there is no change in the individuals (closed model), individuals leave and never return (emigration + mortality), leave and return (emigration + reimmigration), or a combination of the last two (emigration + reimmigration +mortality) (Whitehead, 2001). Model selection was based on the lowest quasi-likelihood Akaike information criterion (QAIC) (Whitehead, 2007). Supported models fell within 2 units (Burnham & Anderson, 2002). Confidence interval and standard error of parameter estimates were calculated using nonparametric bootstrap techniques (100 replicates) (Whitehead, 2007).

| Connectivity
The Satayah and Samadai catalogues were compared to assess the presence of common distinctive individuals in the two populations. The

| Population parameters
The capture histories of Highly Marked Individuals (HMIs, including D1 and D2) in 2010-2013 were pooled in four yearly occasions (2010, 2011, 2012, and 2013) to estimate annual survival and capture probabilities, and population size in program MARK (White & Burnham, 1999). The Global test on the dataset showed overdispersion (χ 2 = 6.994 and p = .14), and CR strict assumptions on capture and survival heterogeneity (due to, among others, transience; see Appendix 1) were tested in UCare (Choquet, Reboulet, Pradel, Gimenez, & Lebreton, 2002). Preliminary analysis of the individual capture histories anticipated the occurrence of transients in the sample. In order to minimize biases on apparent survival (Pradel et al., 1997) and abundance (Pollock et al., 1990)

| Power analyses for population trends
The simplified equation of Gerrodette's inequality model (Gerrodette, 1987), r 2 n 3 ≥ 12CV 2 (Z α/2 + Z β ) 2 , combines information on population rate of change (r), number of estimates available (n), coefficient of variation (CV), and probabilities of Type I (Z α/2, one-tailed) and Type II (Z β ) errors, to calculate how large a trend could have been detected with the data available, and how long a survey would have been required to detect the observed trend. Error probabilities were set to .05 for a 95% power to detect a change (Gerrodette, 1987;Parra, Corkeron, & Marsh, 2006). The probability of making a Type II error (β) was set also to .20 for a more conservative 80% power (Tyne et al., 2016). The overall fractional change in population size and the annual rate of change were calculated with formulae in Gerrodette (1987; Appendix 1) assuming a uniform exponential trend.

| Site fidelity
The best model "emigration + mortality" predicts 42-58 individuals in Satayah Reef at any given time during the study period (2006,(2010)(2011)(2012)(2013), with mean residence times of 2,736 days (approx. 7 years) ( Table 3). The lagged identification rate did not level off above zero at longer time lags, hence excluding residence and/or reimmigration in the site (Whitehead, 2001) (Figure 3). The supported "emigration + reimmigration" model was therefore rejected.

| Power analyses for population trends
The study had a high power (1 − β = .95) to detect a constant rate of change as little as 0.13 per year, which would have resulted in a 34% population decline or 44% increase over the course of the study ( Abbreviations: a, mean residence time (days) in Satayah Reef; b, mean residence time (days) outside Satayah Reef; N, mean population in Satayah Reef at any given time; δ, rate of mortality or permanent emigration (notation follows (Whitehead, 2001).  (Karczmarski et al., 2005;Norris et al., 1994). As well as geographic isolation, social and ecological factors can also have an influence in shaping the structure of insular communities (Oremus et al., 2007). As it cannot be excluded that these Egyptian units connect to each other and/or to larger, Note: Based on Gerrodette's inequality model (1987), with 95% and 80% power, yearly survey intervals (t = 1) and constant coefficient of variation (CV = 0.08).

| D ISCUSS I ON
pelagic populations outside the resting areas, the information available suggests that they could be part of a metapopulation, a structure organized in subpopulations of individuals differentially using a network of habitat patches (Levins, 1969).
The population structure is instrumental in assessing the intrinsic vulnerability of the Satayah population. The exposure of calves and resident individuals to interactions with swimmers and boats, the proven sensitivity (Fumagalli et al., 2018), and individual long-term residence in the study site suggest that this is a vulnerable population, which should be closely monitored to document the occurrence of biological impacts that could be caused, or exacerbated, by the intense WW activities. Such impacts could manifest themselves in two major ways. Firstly, individuals in populations chronically affected by tourism operations and unable to cope with the disturbances may abandon the site and relocate to a less disturbed one (Lusseau, 2004). This is a viable option if alternative suitable sites are available, and the benefits associated with the displacement overcome its risks and costs (e.g., predation, presence of competitors, relations with associates; Frid & Dill, 2002;Gill, Norris, & Sutherland, 2001). However, when this strategy is not advantageous, individuals or groups would continue to use the site despite the disturbances.
This can result in changes in demographic parameters, most likely female reproductive success (Christiansen & Lusseau, 2014), and eventually in decreased population size (Bejder et al., 2006;Lusseau, Slooten, & Currey, 2006). A decline in population abundance was reported from Hawaii, where human interactions with spinner dolphins have intensified over the last few decades .
It is still not clear whether the Egyptian populations are affected by WW interactions, and whether the impacts would lead to displacement or population decline. Furthermore, it must be acknowledged that other phenomena, both natural and anthropogenic (e.g., environmental conditions, resource competition, prey abundance, diseases, overfishing, bycatch) may co-occur, and their effects interact in threatening wild populations. Although, in most cases, it is extremely complex to tear apart the specific effects of single threats, we recommend future studies to maintain a holistic approach and to quantify, describe, and consider all possible sources of disturbance and stress when assessing the status of the Satayah population. As the 2006 study by Bejder and colleagues demonstrates, control-impact studies would be ideal and should be taken into consideration, when possible. Contrasting and comparing resting behavior within and between control and impact resting areas has already advanced the understanding on the short-term effects of disturbances on spinner dolphins in Egypt (Fumagalli et al., 2018). As several resting areas are available to spinner dolphins in Egypt and the Red Sea (Fumagalli, Cesario, & Costa, 2019), efforts should be made to (a) compile pho-toID databases and estimate parameters of the Egyptian populations found at these resting areas, especially those with a potential to be control sites, and to monitor the composition of pelagic schools; (b) quantify and describe the characteristics of WW activities at resting sites, in order to monitor the evolution of the industry and enable the identification of WW variables that may be used in models to measure or predict changes in population demographic parameters (e.g., number of vessels, Bejder et al., 2006;Pérez-Jorge et al., 2016; implementation of regulations, Gormley et al., 2012); and (c) model individual and population temporal and spatial variation in exposure to anthropogenic stressors (e.g., Pirotta et al., 2015).
In this case study, the duration and characteristics of the study did not allow the assessment of population-level impacts.
Nonetheless, three major direct management applications derive from the investigation of the Satayah population ecology. Firstly, the Satayah population is proposed as a management unit. Current knowledge indicates that the Egyptian spinner dolphins are organized in small, discrete units, whose boundaries are still not understood. If the region hosts a metapopulation, adequate site-specific management interventions are required to ensure the viability of each subpopulation (Oremus et al., 2007). Secondly, school demographic composition, individual site fidelity, population size, and survival are suggested as key monitoring indicators. Baseline data are now available for future assessment of impacts and resilience of the population, as well as for inclusion in monitoring frameworks (Higham, Bejder, & Lusseau, 2009). Thirdly, the Satayah population is confirmed ideal for the investigation of population trends, as the conditions that maximize power detection-small, resident, easily accessed population, abundance estimates with good precision (Taylor et al., 2007)-are met. However, careful survey planning and design are required to ensure the best compromise between research effort, objectives, and logistical feasibility for prompt detection of changes. At this scope, simulating scenarios of population change and trend detection could help identify such compromise (e.g., Thompson et al., 2000;Tyne et al., 2016). As the current levels of monitoring are inadequate to detect a population decline in a timely manner, and the local WW industry is on the rise and still unregulated, we strongly support recommendations already made by Tyne and colleagues for the spinner dolphins of Hawaii (Tyne et al., 2016).
We urge the use of a more cautious approach to the management of such industry, including the reliance on a lower power level (80%) and the adoption of precautionary measures to mitigate impacts to, hopefully, help prevent population decline (Tyne et al., 2016).
We caution here that our findings could have reflected seasonal patterns and, given the high proportion of males in the sample, have over-represented sex-specific patterns in residence, which were found at Samadai Reef (Cesario, 2017). Regular surveys throughout the year and a broader temporal and geographic photo-identification effort are required to further resolve the characteristics of Satayah schools, individual residence and dispersal, to compare trends between study sites, and to advance the understanding of the species organization in the region. In future surveys, adjusting the data collection to apply the Robust Design formulation (Kendall, Pollock, & Brownie, 1995;Pollock, 1982), that best accommodates transience and allows estimation of temporary immigration and emigration, is strongly recommended. Given the preliminary evidence of connectivity between the Samadai and Satayah populations, the opportunity to apply Multistate Robust Design models (Kendall & Bjorkland, 2001;Kendall, Nichols, & Hines, 1997;Pollock, 1982;Schwarz & Stobo, 1997) should also be taken into consideration.
Finally, a dedicated survey on the efficacy, precision, implications for capture-recapture analyses and impacts of surface and underwater photoID data collection is required to define whether one, or a combination of the two, provides the best compromise between research needs and dolphin disturbance.

| MANAG EMENT IMPLI C ATI ON S
This study advanced previous knowledge of the potential disruptive nature of WW on this population (Fumagalli et al., 2018)  Conservation, and crowdfunding. We would also like to thank our anonymous reviewers, who provided constructive and critical comments that greatly enhanced the quality of this work.

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R S ' CO NTR I B UTI O N
GNS, MC, and MF conceived the idea and designed methodology; MF, AC, and MC collected the data; MF and AC analyzed the data; ES and JH advised on analyses; MF led the writing of the manuscript.
All authors contributed critically to the drafts and gave final approval for publication.

Photographic quality and distinctiveness
Each image was scored according to four photographic quality criteria (focus, contrast, angle, and fin visibility) adapted from Friday et al. (2000) and Urian et al. (2015), and consistent with Samadai photo-identification studies (Cesario, 2017) (Table A1). The sum of the criteria scores defined excellent, very good, good, fair, and poor photographic quality categories (Table A2).

Capture-Recapture (CR) assumptions
The CR models employed assume that marks do not affect the behavior or fate of individuals (trap response) and are not lost, misread, overlooked, or missed (mark loss); every individual alive at time i has the same probability of capture (equal catchability); the fate of each marked individual is independent of the fate of other marked individuals (independence of fates); no birth, death, immigration, and emigration occur during the resampling process (instantaneous sampling) (Lindberg & Rexstad, 2006). When needed, these assumptions can be relaxed, accommodated, or corrected (White & Burnham, 1999).
To enhance validation, we adopted the methods and strategies described in Table A3. 95CI N = 95% confidence interval of N (Burnham et al., 1987;Williams, Nichols, et al., 2002). F I G U R E A 2 Individual distinctiveness: example of notch, nick, small nick, and tick

Power analyses for population trends
A i = abundance at occasion i; A 1 = initial abundance; CV = coefficient of variation; n = the number of samples; r = fractional rate of change of the quantity being measured; α and β, the probabilities of Type 1 and 2 errors, 2 res = estimate of residual variance (Gerrodette, 1987 (Pollock et al., 1990;Pradel, 1993;Williams, Trites, & Bain, 2002) Mark loss and recognition Marks are not lost, missed, overlooked or misread Data processing: Highly marked individuals only; High-quality pictures (Barlow et al., 2011;Frasier, Hamilton, Brown, Kraus, & White, 2009); Experienced cataloguer (Pollock et al., 1990;Williams, Nichols, et al., 2002) Equal catchability Every marked individual alive in the population at time i has the same probability of capture Pooled chisquared statistics (Test 2 + Test 3) Survey design: Area surveyed correspond with home range; Seasonal phenomena that may affect individuals' presence are taken into consideration (Hines, Kendall, & Nichols, 2003). Data collection: Even coverage of groups

Independence of fates
The fate of each marked individual is independent of the fate of other marked individuals Data processing: Exclude individuals not mixing at random (e.g., calves) (Rosel et al., 2011) Instantaneous sampling Resampling is instantaneous; that is, birth, death, immigration, and emigration do not occur during the resampling process Survey design: Sampling occasions are short in duration (Pollock et al., 1990;Williams, Nichols, et al., 2002)