Implications of survey effort on estimating demographic parameters of a long‐lived marine top predator

Abstract Effective management of wildlife populations rely on knowledge of their abundance, survival, and reproductive rates. Maintaining long‐term studies capable of estimating demographic parameters for long‐lived, slow‐reproducing species is challenging. Insights into the effects of research intensity on the statistical power to estimate demographic parameters are limited. Here, we investigate implications of survey effort on estimating abundance, home range sizes, and reproductive output of Indo‐Pacific bottlenose dolphins (Tursiops aduncus), using a 3‐year subsample of a long‐term, capture–recapture study off Bunbury, Western Australia. Photo‐identification on individual dolphins was collected following Pollock's Robust Design, where seasons were defined as “primary periods”, each consisting of multiple “secondary periods.” The full dataset consisted of 12 primary periods and 72 secondary periods, resulting in the study area being surveyed 24 times/year. We simulated reduced survey effort by randomly removing one, two, or three secondary periods per primary period. Capture–recapture models were used to assess the effect of survey intensity on the power to detect trends in population abundance, while individual dolphin sighting histories were used to assess the ability to conduct home range analyses. We used sighting records of adult females and their calving histories to assess survey effort on quantifying reproductive output. A 50% reduction in survey effort resulted in (a) up to a 36% decline in population abundance at the time of detection; (b) a reduced ability to estimate home range sizes, by increasing the time for individuals to be sighted on ≥30 occasions (an often‐used metric for home range analyses) from 7.74 to 14.32 years; and (c) 33%, 24%, and 33% of annual calving events across three years going undocumented, respectively. Results clearly illustrate the importance of survey effort on the ability to assess demographic parameters with clear implications for population viability analyses, population forecasting, and conservation efforts to manage human–wildlife interactions.

Demographic parameters vary between populations of the same species and between species (see McMahon, Burton, & Bester, 2003;Moss, 2001;Nicholson, Bejder, Allen, Krützen, & Pollock, 2012;Sprogis et al., 2016a;Wittemyer et al., 2013). Inferring demographic parameters estimated for one population to another population of a similar species may result in inaccurate conclusions when evaluating their long-term viability (Manlik et al., 2016). The time required to accurately assess demographic parameters of long-lived, slowly reproducing species can take years (Mann, Connor, Barre, & Heithaus, 2000;Moss, 2001). Optimizing the power of a survey design to meet its research objectives is therefore critical (Hawkins et al., 2017). While several monitoring studies have quantified the power of survey designs to detect trends in population abundance (e.g., Ansmann, Lanyon, Seddon, & Parra, 2013;Brown, Bejder, Pollock, & Allen, 2015;Parra, Corkeron, & Marsh, 2006;Tyne et al., 2016;Wilson, Hammond, & Thompson, 1999), they have ignored the potential implications of survey design for the assessment of additional demographic parameters such as reproductive and survival rates.
The waters off Bunbury, Western Australia, are home to a resident population of Indo-Pacific bottlenose dolphins (Tursiops aduncus). Abundance estimates vary seasonally from a minimum of 76 dolphins (95% CI 68 to 85) in the winter to a maximum of 185 dolphins (95% CI 171 to 199) in the summer (Sprogis et al., 2016a). Here, individuals utilize the area differently, with animals sighted in open waters having larger home ranges than those utilizing predominately inshore waterways (Sprogis, Smith, Rankin, MacLeod, & Bejder, 2016b). The Port of Bunbury is currently the fourth largest port in Western Australia (Ports Australia 2013) and is expanding its capacity to a greater extent to support growing recreational and commercial vessel operations (Australian Government Department of the Environment 2016; Taylor Burrell Barnett 2015). In addition, dolphin-based tourism (including boat-based dolphin eco-cruises, swim-withdolphin tours, and licensed food provisioning of wild dolphins) represents a substantial proportion of Bunbury's tourism economy (Smith, 2012). Consequently, the Bunbury dolphin population is exposed to multiple sources of human activities, including recreational vessel traffic (Jensen et al., 2009), commercial shipping traffic, commercial tourism, and both licensed (legal) and unlicensed (illegal) food provisioning (Arcangeli & Crosti, 2009;Smith, 2012).
The long-term viability of the Bunbury dolphin population is projected to decline by 50% in the next 20 years (Manlik et al., 2016). Low reproductive rates have been identified as the leading cause for the decline (Manlik et al., 2016). Historically, the Bunbury dolphin population has served as a "source" population for the larger meta-population along the southwestern Australian coast (Manlik et al., In Press), and more recently, the abundance and temporary emigration of the population are also influenced by climate variability (Sprogis, Christiansen, Wandres, & Bejder, 2018), thus raising additional conservation concerns for the overall meta-population viability. Therefore, it is important to understand both the ability to detect trends in population abundance and to quantify demographic parameters for this population to best inform management.
Here, we utilized a 3-year subsample of an ongoing longterm capture-recapture photo-identification study focussing on the local dolphin population (Smith, Pollock, Waples, Bradley, & Bejder, 2013;Sprogis et al., 2016aSprogis et al., , 2018. This study conformed to the structure of Pollock's Robust Design (hereafter referred to as "Robust Design"; Pollock, 1982). The objectives were to assess how various levels of survey effort impacted our ability to (a) detect trends in population abundance; (b) quantify apparent survival rates; (c) conduct home range size analyses; (d) detect calving events; and (e) quantify the uncertainty surrounding a calf's period of birth. This study provides novel insights into the implications of survey effort to estimate several key demographic parameters of a long-lived species.

| Data collection
Boat-based photo-identification surveys for dolphins have been aimed to photograph every dolphin's dorsal fin for identification (Würsig & Jefferson, 1990). A group was defined as one or more dolphins within 100 m of any other dolphin and involved in the same or similar behavioral activity (Smith et al., 2013). For further details on data collection and study design, see Smith et al. (2013) and Sprogis et al. (2016a).

| Capture-recapture sampling design
Boat-based surveys conformed to the Pollock's Robust Design capture-recapture method (Pollock, 1982;Sprogis et al., 2016a). The Robust Design model is constructed of a series of "primary sampling periods" (hereafter referred to as primary periods), each consisting of multiple "secondary sampling periods" (hereafter referred to as secondary periods). A population is assumed to be open between primary periods and closed within each primary period. In Bunbury, each primary period consisted of one austral season of sampling effort: summer (December-February); autumn (March-May); winter (June-August); or spring (September-November). Each season consisted of six secondary periods. The completion of the three transect zones (Buffalo Beach, Back Beach, and Inner waters) defined a secondary period (Figure 1). Sprogis et al. (2016a) defined the Robust Design model assumptions and the steps taken to minimize violations of these assumptions in this study. As such, the full data set consisted of 12 primary sampling periods equivalent to 72 secondary sampling periods over a 3-year period, resulting in the study area being surveyed 24 times per year.

| Data processing
In the photographic images of dolphin dorsal fins, unique nick and notch outlines were used to identify each individual (Wursig & Wursig, 1977) to a long-term catalogue. Two or more researchers independently conducted the fin-matching process for each indi-

| Survey sampling scenarios
We explored our ability to detect trends in population abundance, estimate apparent survival rates (i.e., the total effect including true survival and emigration), conduct home range analyses, detect reproductive events and the precision of new calves ages for four scenarios of survey effort over 3 years. Specifically: • Original Data: The original capture history for the 12 primary periods sampled, each consisting of six secondary periods (i.e., the entire study area being surveyed 24 times per year for 3 years).
• Simulation 1: A simulated reduction of survey effort to five secondary periods per primary period, by removing one secondary period per primary period.
• Simulation 2: A simulated reduction of survey effort to four secondary periods per primary period, by removing two secondary periods per primary period.
• Simulation 3: A simulated reduction of survey effort to three secondary periods per primary period, by removing three secondary periods per primary period.

F I G U R E 1
The 120-km 2 study area off Bunbury, Western Australia. The study area was divided into three transects (dashed lines) along which boat-based photo-identification capture-recapture surveys were conducted for Indo-Pacific bottlenose dolphins: Buffalo Beach, Back Beach, and Inner waters transects Further reduction of survey effort was not explored due to model limitations. Secondary periods were randomly removed and the process was repeated 100 times for simulations 1 to 3. All analyses and modeling were conducted using R 3.1.2 software package (R Core Team 2014) unless otherwise noted.

| Detecting trends in population abundance and estimating apparent survival rates
Each of the capture history datasets applied Robust Design capture-recapture models with the RMark interface (Laake, 2013).
RMark provides an R-based interface linked to the program Mark (White & Burnham, 1999). Based upon previous modeling work (Sprogis et al., 2016a), models were fit to constant survival (φ(.)), time-varying Markovian emigration (γ″(t) ≠ γ′(t)), and time-varying capture probabilities within primary periods (p = c(t,s)) (e.g., the best-fitting model for adults and juveniles was φ(.) γ″(t) ≠ γ′(t) p = c(t,s)). The coefficient of variation (CV) and survival rate estimation for each model run were retained. The average CV for each scenario was calculated and used in analysis for detecting trends in population abundance.
Gerrodette's inequality model (Gerrodette, 1987) was applied following the method presented in Tyne et al. (2016). For each of the four scenarios, we used the average CV using the software package Trends (Gerrodette, 1993) to quantify the time required under each scenario to detect a 5% and 10% change in abundance at a statistical power of 0.8 and 0.95.

| Quantifying effect of modified survey effort on the ability to conduct home range analyses
We used four metrics to explore the effect of survey effort on the abil-

| Quantifying effects of reduced survey effort on documenting calving events
We quantified the effect of reduced survey effort on our ability to detect calving events in the first year of life. We utilized sighting histories containing for reproductively active females during this study (December 2009to November 2012. When a female was photographed with her dependent calf, the calf's presence was added to the sighting history. Following the same procedure used for the capture-recapture modeling, we then simulated reduced survey effort with one, two, or three secondary periods removed from each primary period (simulations 1-3 respectively). This process was TA B L E 1 Number of years to detect change in population abundance, percent decline/increase at the time of detection at two annual rates of change (0.05 and 0.1) at power = 95% or power = 80% with four (seasonal) abundance estimates per year Note. Original Data = six original secondary periods, Simulation 1 = five randomly subsampled secondary periods, Simulation 2 = four randomly subsampled secondary periods, and Simulation 3 = three randomly subsampled secondary periods.
repeated 100 times for each simulation of survey effort. From the Original Data set and for each simulation, the number of calves documented in the year in which they were born was retained.

| Ability to detect changes in abundance based on survey effort
Under the Original Data, the precision of abundance estimates was high with an average CV (precision) of 0.05. Reduction of survey effort by one, two, or three secondary periods increased the average CV to 0.07, 0.08, and 0.12, respectively (Table 1; Figure 2). Detection of a 5% annual change in population abundance at 80% power would take 2.75-5.5 years (Original Data and Simulation 3, respectively) and 3.25-6.5 years at 95% power (Original Data and Simulation 3, respectively;

| Estimation of apparent survival rates based on survey effort
The

| Ability to conduct home range analyses based on survey effort
Thirteen individuals were seen on ≥30 occasions in the 3-year study period of the Original Data set. Under all scenarios, no individual was observed on ≥50 occasions throughout the 3-year period (Table 2). Reducing survey effort by one secondary period per primary period (Simulation 1) resulted in a decrease to 5.55 individuals (average) sighted ≥30 times (Table 2). When survey effort was reduced by half (Simulation 3), no individual was seen on ≥30 occasions (Table 2).
Average sighting frequency during the 3-year study period ranged from a minimum of 6.29 (±0.25 SD) sightings under Simulation 3 to a maximum of 11.63 sightings within the Original Data. Based on these results, it would take between 7.74 (Original Data) and 14.33 years (Simulation 3) for an individual, identified at the average sighting frequency, to be sighted ≥30 occasions (Figure 4). For an individual that was identified at the average sighting frequency, it would take 12.90 (Original Data) to 23.88 years (Simulation 3; Figure 4) to be seen ≥50 times.  Figure 5). In contrast, Simulation 3 resulted in, an average, 32.8%, 23.7%, and 33.1% of calves undocumented in the year they were born (Table 3; Figure 5).

| Ability to detect changes in abundance under current and alternative survey effort
The ability to detect trends in abundance of marine mammals is limited in many cases. Except for pinnipeds, marine mammal monitoring programs have not been able to detect population declines in ≥50% of all taxonomic groups examined (Taylor, Martinez, Gerrodette, Barlow, & Hrovat, 2007). Compared to other studies (e.g., Brown et al., 2015;Parra et al., 2006;Tyne et al., 2016;see  Note. Original Data contained the data set consisting of six secondary periods per primary period, while simulations 1, 2, and 3 were simulated reduced survey effort by one, two, and three secondary periods per primary period, respectively.  effort (Original Data). When survey effort was reduced by half (Simulation 3), 6.5 years were required to detect a 5% change in population abundance at 95% power. Previous studies found similar time periods required, with 7 years for Hawaiian spinner dolphins (Stenella longirostris; Tyne et al., 2016), 4 or 10 years for Indo-Pacific bottlenose dolphins (Ansmann et al., 2013), 6 years for snubfin dolphins (Orcaella heinsohni; Parra et al., 2006), and 10 years for humpback dolphins (Sousa sahulensis; Parra et al., 2006; Table 4). Findings presented in this study suggest that reducing survey effort by half (Simulation 3) could result in a population abundance decline of 36% at the time of detection. The lack of ability to readily detect significant abundance decline significantly hinders the implementation of effective management efforts.

| Effects of reduced survey effort on estimates of apparent survival
The apparent survival rates estimate here are similar to those documented for other free-ranging dolphin populations (e.g., Brown et al., 2015;Ryan, Dove, Trujillo, & Doherty, 2011;Tezanos-Pinto et al., 2013;Tyne, Pollock, Johnston, & Bejder, 2014), including those previously published for the Bunbury dolphin population (Smith et al., 2013;Sprogis et al., 2016a). Across all scenarios, apparent annual survival rates ranged from 0.949 to 1, with little difference as survey effort decreased. One possible explanation for this is that the original apparent survival rates were high due to the structure of the capture history and resighting rates of individuals (Original Data; 0.98); consequently, the removal of secondary periods resulted in estimates being relatively insensitive to the simulated reduction of survey effort. A reduction of survey effort was unlikely to alter the conclusions drawn pertaining to apparent annual survival rates for the population or to trigger alternative management approaches on its own. However, when combined with other demographic parameters (such as calf survival rates or fecundity estimates), a decrease in the estimated apparent survival rate may alter model forecasts in population viability analyses (Currey, Dawson, & Slooten, 2009;Manlik et al., 2016).

| Implications for conducting home range analyses
Our results highlighted two potential limitations of reducing survey effort for conducting home range analyses. Previous research highlighted the importance of having ≥30 independent sightings of an individual (and preferable ≥50 sightings) to accurately estimate its home range size (Seaman et al., 1999). Our findings suggest that reducing survey effort from six secondary sampling periods per primary period (Original Data) to three secondary sampling periods (Simulation 3) increased the time required for an individual to be sighted on ≥30 occasions from 7.74 years (Original Data) TA B L E 4 Overview of delphinid studies that assessed the ability to detect changes in population abundance, with the coefficient of variation (CV) and the number of years to detect a change in abundance displayed  (1999) to nearly 15 years (Simulation 3; Figure 4). Similarly, the time for an individual to be observed on ≥50 occasions increased from an average of 12.90 years (Original Data) to 23.88 years (Simulation 3; Figure 4). A natural consequence of the inability of a sampling design to exceed a minimum individual sighting frequency is the concurrent decline in total sample size for analyses (i.e., in the case the number of individuals for which home range size can be estimated). An increased number of years required to gather sufficient data to conduct home range analyses would eliminate our ability to examine whether individuals alter their usage of an area as the result of perturbation in their environment (Börger et al., 2006;Sprogis et al., 2016b).
The weaning age of bottlenose dolphins (Tursiops sp.) is typically between ages of 3 and 6 years (Mann et al., 2000;Wells, 2014), with the majority weaned by 4 years of age (Mann et al., 2000).
For some bottlenose dolphin populations, first-year calf mortality ranges between 15% and 42% (Mann et al., 2000;Wells, 2014), with over 40% of calves not surviving to weaning (Mann et al., 2000). This highlights that a female dolphin could give birth to a calf and the calf could die before being documented via the research monitoring program. Our simulations quantified the effect of reduced survey effort on our ability to determine the timing of new calving events. Short-time intervals between successive surveys will optimize documentation of calving events. Year-round survey effort at reduced levels of survey effort was not enough to ensure that all calving events were documented in the first year of birth. Our results highlight that sustained high levels of year-round survey effort were required, as reduced survey effort resulted in 5%-40% of calving events not being documented in the year during which they occurred.
Undocumented calving events will have multiple effects on estimates of demographic parameters. For example, intercalving intervals will be positively biased (Mann et al., 2000), resulting in lower fecundity or reproductive output estimates (Arso Civil, Cheney, Quick, Thompson, & Hammond, 2017). Missed calving events would also positively bias calf survival rates as the true calf mortality rate would be higher than that estimated (Mann et al., 2000).
Consequently, utilizing parameters estimated from reduced survey effort are likely to result in inaccurate forecasts being made about the long-term viability of a population (Currey et al., 2009;Manlik et al., 2016). The impacts of which can reduce the ability of a longterm dataset to accurately inform management and policy decisions (Hughes et al., 2017).

| Applicability of methods to other study systems
Long-term monitoring programs serve a critical role in informing wildlife policymaking and our understanding of ecological systems (Editorial 2017;Hughes et al., 2017;Mann & Karniski, 2017). Such programs are costly (Tyne et al., 2016), and securing funding is an ongoing challenge (Editorial 2017; Hughes et al., 2017). Consequently, understanding the potential implications of reduced survey effort resulting from limited resources on the ability to estimate population demographic parameters is necessary. The methods presented here are broadly applicable to a wide range of study systems. The ability to detect changes in population abundance is applicable for any study system where the precision (CV) of abundance estimates can be quantified (Ham & Pearsons, 2000;Johnson, Camp, Brinck, & Banko, 2006;Taylor et al., 2007) can be used to assess the implications of reduced survey effort for conducting home range analyses. Lastly, in study systems with uniquely identifiable individuals where parental care is provided to young (e.g., birds and mammals), the potential consequences of reduced survey effort on the ability to detect reproductive events can be quantified using the methods applied in this study.

| CON CLUS IONS
Our results highlight the effect of survey intensity on estimating abundance and demographic parameters of a long-lived marine top predator. All four survey effort scenarios tested in this study are relatively high in sampling frequency (