Under Pressure: Comparing Fish Survivorship Between in Situ and Boat Tagging Methods Using Internal Acoustic Tags


 Background: With the increase in telemetry studies over the past decade, improving understanding of how different tagging methods influence fish survivorship is critical. By examining the effects of tagging methods, we can maximize the information gained from telemetry studies. Mortality resulting from internally tagging fish on a boat may be due to barotrauma injuries, increased stress from prolonged handling times, or predation after fish have been released back into the water. Conducting in situ internal acoustic tagging at depth of capture completely removes barotrauma stresses and simplifies the release method, which may improve fish survival. In this study, we used 8 years of acoustic tagging data to determine if the tagging method (in situ versus on the boat) influenced fish survivorship and evaluated the role of other tagging variables.Results: At 6 days after tagging, Kaplan–Meier survival curves revealed that the survival probability of fish tagged on the boat was 66% while survival probability of fish tagged in situ was 90%. Tagging method was the only variable to significantly affect survival probability based on Cox proportional hazards models, with fish tagged in situ ~75% less likely to have an “event” (mortality, tag loss, or emigration) compared to fish tagged on the boat at both 4 and 6 days after tagging. Examining tagging methods separately, handling time only marginally influenced survival probability of boat-tagged fish and no variables had a significant effect on survival of in situ tagged fish. Conclusions: In this study, tagging method was the only variable to significantly affect survival of internally tagged fish. Implanting internal acoustic tags in situ is not a practical method for every species and for every environment, but given the increased fish survivorship demonstrated here, we strongly suggest it be considered as the preferred tagging methodology where applicable.


Background
Acoustic telemetry has become a widely accepted method for collecting animal movement data in the marine environment . Data collected from acoustically tagged sh can be used to investigate a wide variety of biological and ecological questions. For example, acoustic telemetry data have provided answers to broad ecological questions, such as revealing previously unknown migration patterns for certain species (Feeley et al. 2018, Pratt et al. 2018) and have also be used to answer more localized questions, such as patterns of habitat use in a speci c location (Herbig et al. 2019, Keller et al. in press). Although acoustic telemetry studies have been rapidly increasing in number over the past 20 years (Crossin et al. 2017), there are far fewer studies that examine the in uence of tagging methodology on acoustic telemetry results (Dance et al. 2016). This information is needed because the physical act of tagging a sh can affect the outcome of the tagging event, potentially change the behavior of that sh, and even in uence the interpretation of results.
In traditional mark/recapture studies, external tags are inserted into the dorsal musculature. Traditional mark/recapture studies are a cost-effective method of tagging animals, but these studies may provide only limited spatial information, require the recapture of tagged sh, and often have low success rates (Lucas and Baras, 2000). To overcome some of these limitations, researchers use more advanced tags, such as acoustic tags. While externally attaching acoustic tags has been accomplished through a wide array of techniques, these tags can be quickly shed, causing a truncation of data collection (Jepsen et al. 2015). An alternative method of acoustic tagging that increases tag retention is internally inserting the tag, usually through surgery (Wagner et al. 2011). Mortality during the process of internally tagging a sh on a boat may be due to stress from barotrauma, changes in water and air temperature, prolonged sh handling, or the amount of time the sh spends on the surface (Williams et al. 2015). Even with survival from the tagging surgery, there is still the potential for physical trauma (bruising, bleeding, or acute damage) or physiological disturbance (changes affecting gasses, blood, and pH) associated with stress from capture or mishandling (Skomal 2007, Gallagher et al. 2014, Wilson et al. 2014).
Additionally, newly tagged sh may be more susceptible to predation, especially if the sh must travel through the water column before reaching a protective habitat (Piraino and Szedlmayer 2014). Acoustic tagging release method experiments have demonstrated that sh descender devices, which return sh to depth, can increase sh survivorship (Bohaboy et al. 2019), but devices that both return a sh to depth and protect against predators (e.g., cages) are best for increased sh survivorship (Piraino and Szedlmayer 2014; Williams et al. 2015). However, a review of barotrauma treatment (venting) in catchand-release studies revealed inconsistency about which method is most bene cial in reducing or treating barotrauma, likely because of differences in environmental or species-speci c physiological variables and differences in assessment methods (Eberts and Somers 2017).
All tagging methods have the potential to cause health issues for the sh, so decreasing the risk of health issues should be a priority when planning a tagging project. In the case of acoustic telemetry, it is often di cult to determine the fate of sh when they are no longer detected. While effects of the tagging procedure can cause mortality through physical trauma, the sh may also experience mortality via predation or shing pressure, lose the transmitter, or immediately emigrate from the area. A tagging procedure that minimizes stress to sh may also minimize the risk of these events by shortening the surgery recovery period and reducing the risk of both predation and emigration by releasing the sh directly into the natural sheltering habitat.
We suggest consideration of conducting in situ internal acoustic tagging of sh at depth of capture to completely remove barotrauma stressors and simplify the release method. Tagging sh in situ with acoustic transmitters at the same depth as capture and release has become more popular as studies have shown the negative effects of barotrauma on sh survival (Rummer and  While there has been comparison of discard mortality between in situ and boat tagged sh using external tags (Rudershausen et al. 2014), to our knowledge there has been no study directly comparing survivorship between boat and in situ tagging methods when using internal acoustic tags. In this study, we use 8 years of acoustic tagging data and our personal experiences with surgically implanting transmitters in sh to examine 1) the in uence of tagging method (in situ versus on the boat) on sh survivorship and 2) the effects of other tagging variables (e.g., handling time and surgeon experience) that might also play a role. . The surgical procedures during boat and in situ tagging were similar with the exception that anesthesia (AQUI-S, 50% isoeugenol; aqui-s.com) was not used in situ. We noticed that sh were calm once they were turned ventral side up and their behavior was similar to that of sh that had been anesthetized. Tonic immobility is commonly used as an alternative to chemical-based anesthesia for surgical implantation in elasmobranchs Hussey 2015, Sloan et al. 2019) and after a few initial efforts, we realized that use of chemical-based anesthesia for tagging sh in situ was not necessary or bene cial and did not allow a quick and immediate release of the sh after surgery was completed. During both tagging methods, the following tagging variables were recorded: species, total length of sh (TL), depth of capture, tagging location, handling time, tagging time, and name of the person performing the surgery.

Fish tagging
Fish that did not have complete tagging procedure data were excluded from analyses.

Data validation
To validate detection data, sh with only one detection were excluded because a single detection could be caused by collisions or noise and is not considered reliable (Simpfendorfer et al. 2015). Additionally, detections from sh that were detected only on one receiver were examined to determine whether detections were the result of normal sh activity or from a tag loss or mortality event. Consistent detections regardless of hour or day or a pattern of higher detections during a particular diel period could be due to a tag sitting on the sea oor with changes in environmental conditions throughout the day in uencing transmission success.

Tagging variables
Tagging time was the total time of the tag implantation surgery, while handling time was either the total time the sh was on the boat or total time the sh was handled underwater. "Fight" times for sh caught via hook and line were not recorded, so analyses examined only the amount of time the sh was on the boat. Tagging location was the general location of the tagging, either the Lower Florida Keys or the Dry Tortugas ( Fig. 1), and was included to account for any environmental differences between locations. The family name of sh (n=2), either Serranidae or Lutjanidae, was also included as a covariate to examine differences in survival based on physiological differences between taxonomic groups but to avoid overparametizing the models because of the high number of species (n=14).
Total length of sh at time of capture was transformed into a two-factor variable called " sh size" to compare small and large sh without the confounding factor of sh family. Mean total lengths were determined separately for each family, and sh were designated as "small" if their length was below this value for the respective family and "large" if their length was equal to or greater than this value.
To determine if the tagging experience of the surgeon had an in uence on survivorship, a variable called "surgeon experience" was created. The name of the surgeon and date and time of the surgery were used to assign each sh a surgery value based on how many surgeries the surgeon had previously completed.
For example, surgery had a value of one if it was the rst sh the surgeon had ever tagged or a value of 17 if it was the 17 th sh a surgeon had tagged. If that value was less than the mean number of surgeries completed (among all surgeons), the surgeon experience was assigned as "low." Alternatively, if the value was higher than the mean number of surgeries completed, the surgeon experience was assigned as "high."

Survival analysis
We examined survivorship of tagged sh using a Cox proportional hazards model, which is a common shorter time frame may miss delayed effects from tagging and longer time frames increase the chance of mistaking behavioral events for those related to the tagging procedure. Due to discrepancy in time after tagging to analysis of survival from the tagging procedure, we decided to determine survival at both 4 and 6 days after tagging. For each sh, the last day of its detection was used to calculate the number of days it survived before an event. For example, if a sh was last detected 5 days after the tagging procedure, it was marked as having an event at day 5. The term "survival" is used to refer to the sh being present and detected by a receiver while an "event" could be mortality via predation or shing pressure, tag loss, or emigration. If a sh was detected past the 4-or 6-day mark, the number of days it survived was truncated to 4 or 6 days, depending on which time frame was used for analysis.
To visually compare the survival probability and number of events between the two tagging methods, Kaplan-Meier survival curves were created using the "survival" package (Therneau 2015) in R (R Core Team 2019). To examine what factors were most in uential on survival probability, Cox proportional hazards models were calculated at both 4 and 6 days after tagging, also using the "survival" package.
Model selection started with the full model using all covariates that were not correlated with tagging method (the main variable of interest) or each other. The best-tting model was chosen through a backwards stepwise search, starting with the full model, and model selection based on lowest Akaike information criterion with a correction for small sample size (AICc) using the "MASS" package (Venables and Ripley 2002). Beyond the initial Cox proportional hazards model examining the in uence of tagging method on sh survival, two secondary hazards models were performed to assess which variables in uenced survival for 1) sh tagged on the boat and 2) sh tagged in situ.
Hazard ratios from the hazards model provide information on the covariates' effects on survival. Hazard ratios above a value of 1 indicate an increase in risk of an event compared to a reference point, whereas a value below 1 indicates a reduction in risk. For categorical covariates, a hazard ratio less than 1 is the proportional reduction in risk compared to the reference level and vice versa. For continuous covariates, a hazard ratio less than 1 is the proportional reduction in risk of an event as the value of the covariate increases by 1 unit. A hazard ratio greater than 1 is the proportional increase in risk of an event as the value of the covariate increases by 1 unit. The farther away a hazard ratio is from 1, the greater the in uence of that covariate. A hazard ratio near 1 means that covariate has a marginal effect on the risk of an event (and thus probability of survival) while a hazard ratio at 1, or a con dence interval that crosses 1, has no effect on the risk of an event (Kleinbaum and Klein 2012).
Independence between residuals and time is an assumption of the proportional hazards model (Kleinbaum and Klein 2012) and was checked in all models by testing the correlation between scaled Schoenfeld residuals and time using the "cox.zph" function in the "survival" package. A signi cant correlation of Schoenfeld residuals and time rejects the null hypothesis and violates the assumption of the proportional hazards model.

Results
Of the sh tagged from 2008-2016, 194 were internally tagged grouper and snapper. Fish with incomplete tagging information were excluded and two sh were excluded from analyses because they had only a single detection. No tagged sh detected at only one location had continuous, consistent detection patterns suspected of resulting from a discarded tag. Additionally, receivers in both arrays were often spaced farther apart to try to maximize coverage area. Most sh detected at a single location were groupers or small snappers that were not expected to make long-distance movements, and smaller movements could have been missed because of spacing between receivers. In total, 105 sh from 14 different species were included in analyses (35 boat tagged and 70 in situ tagged; Table 1).
The average total length was 57.0 ± 1.4 cm (mean ± SE) for Lutjanidae and 64.6 ± 1.7 cm for Serranidae. These values were used to determine whether the variable of " sh size" was "small" or "large" The average number of surgeries completed per surgeon was 9.55 ± 2.92, and this value was used for the determination of whether the variable "surgeon experience" was "low" or "high." Mean handling time and depth varied between tagging method, but mean tagging time was similar ( Table 2).
In situ tagged sh had a higher probability of survival (and remaining in the area) and lower chance of experiencing an event compared to sh tagged on the boat, based on Kaplan-Meier survival curves (Fig. 2). Six days after tagging, the survival probability of sh tagged on the boat was only 66% while survival of sh tagged in situ was 90%. The null hypothesis that there was no difference between the two methods in the probability of an event (mortality, tag loss, or emigration) from day of tagging to 6 days after tagging was rejected by a log-rank test (p = 0.002).   Fish size and tagging time were the only variables not signi cantly correlated with tagging method (the main variable of interest) or each other and therefore were the only variables included in the initial Cox proportional hazards models (Fig. 3). Variables that were signi cantly correlated with tagging method or with each other could not be included in the hazards model without compromising model results.
Because of correlation among variables, model selection using backward stepwise selection started with a full model including only tagging method, tagging time, and sh size. Backward stepwise model selection found the best tting hazards model (based on lowest AICc) at the 4-day mark to include tagging method and tagging time as covariates, while at the 6-day mark, tagging method was the only variable in the best tting model (Table 3).  Hazards model results for the best tting models indicate that sh internally tagged in situ were 76% less likely to have an event (one minus the hazard ratio) 4 days after tagging and 75% less likely to have an event 6 days after tagging compared to sh tagged on the boat (Table 4). Tagging time was also included in the best-tting 4-day model, though the 95% con dence interval spanned 1, indicating there was no effect. For all covariates in both best-tting models, the proportional hazard assumption was supported by a nonsigni cant relationship between scaled Schoenfeld residuals and time. To visually compare each covariate's risk to sh survival, hazard ratio forest plots were produced from the full and best-tting models, and plots for 4 days after tagging are shown as an example (Fig. 4a and 4b). Table 4 Cox proportional hazards models results for the best-tting model at 4 and 6 days after tagging. For the categorical covariates, the hazard ratio is the proportion of risk compared to the reference covariate. For the continuous covariates, the hazard ratio is the proportion of risk with an increase of 1 unit (e.g., 1 min).
A variable with a hazard ratio close to a value of 1 has a marginal effect and a con dence interval (CI) spanning 1 has no effect on hazard. All models include survival as the dependent variable and list independent variables included in the best-tting models. P-values are from the log-rank test.  Pearson correlation coe cients were also calculated for all covariates when splitting the data into sh tagged on the boat and sh tagged in situ. Handling time and depth were the two main covariates of interest. However, since these two covariates were correlated for sh tagged on the boat (Fig. 5a), they could not be included in the same hazards model. Fish size and surgeon experience (thought to secondarily in uence survival) were also correlated with each other. As a result of these correlations, three full models were created for sh tagged on the boat: 1) one with depth, surgeon experience, and all variables not correlated with each other, 2) one with depth, sh size, and all variables not correlated with each other, and 3) one with handling time, surgeon experience, and all variables not correlated with each other. Handling time was correlated with sh size, and thus the two could not be included in the same full model.
For in situ tagged sh, handling time and depth were not correlated with each other, but depth was correlated with sh size (Fig. 5b). Two full models were created for in situ tagged sh: 1) one with handling time, depth, and all other variables not correlated with each other, and 2) one with handling time, sh size, and all other variables not correlated with each other. Handling time was the only covariate in the best tting model for boat-tagged sh at 4 days after tagging (Table 5). At 6 days after tagging, both handling time and family were included in the best-tting model. Handling time was signi cant in both best-tting models; however, the hazard ratio was very close to 1, meaning the effect of handling time on hazard of an event was marginal (Table 6). A hazard ratio of 2.55 for family Serranidae in the 6-day model indicated an increase in risk compared to Lutjanidae. However, this covariate had a wide con dence interval that crossed 1, indicating no signi cant effect of hazard and a somewhat unreliable point estimate (Kleinbaum and Klein 2012). For all covariates in both best-tting models, the proportional hazard assumption was supported by a nonsigni cant relationship between scaled Schoenfeld residuals and time. For in situ tagged sh, handling time was the only covariate in the best tting model at 4 days after tagging and sh size was the only covariate in the best-tting model at 6 days after tagging (Table 7).
However, the global log-rank test examining signi cance of the model overall was nonsigni cant in both cases. Additionally, the con dence intervals for covariates in both models crossed 1, indicating no effect, and the particularly wide con dence interval for sh size in the best-tting 6-day model indicated the point estimate was unreliable (Table 8). For all covariates in both best-tting models, the proportional hazard assumption was supported by a nonsigni cant relationship between scaled Schoenfeld residuals and time.

Discussion
Using an historic dataset of 8 years of sh tagging by Florida Fish and Wildlife Conservation Commission scientists in the Florida Keys, we examined factors a posteriori that in uence survivorship of internally tagged sh. At 6 days after tagging, Kaplan-Meier survival curves demonstrated that sh tagged in situ had a 24% higher survival probability than sh tagged on the boat. Tagging method was the only factor affecting the risk of an event when we examined the initial Cox proportional hazards model assessing survival of all sh. Tagging in situ reduced the chance of sh having an event and thus increased the probability of survivorship by ~ 75% compared to tagging on the boat.
In this study, there were several variables that could not be included in the initial hazards model because of correlation, but their inclusion and limited in uence in the two subsequent models assessing hazards to the boat-tagged and in situ tagged sh independently reinforced that the tagging method was the most important factor for survivorship. Acoustic tracking data of in situ surgically tagged sh from two other studies also indicated high survival rates (97-100%) post-surgery (Tuohy et al divers after they recovered from anesthesia to ensure a safe return to the reef. Nevertheless, the tagged sh may still have suffered from barotrauma, increased stress from being caught by hook and line, or long-lasting effects of anesthesia. All these factors could increase risk of boat-tagged sh having an event. Unfortunately, it was not possible to know the exact fate of sh that had an event because this study did not include a VEMCO Positioning System (VPS) for triangulation of exact location or predation tags that change the transmitter ID code when a tag has entered the stomach of a predator.
At the start of in situ tagging, we decided not to use anesthesia at depth because we discovered that sh held ventral side up with their eyes covered were just as calm as anesthetized sh. We also deemed it advantageous to avoid unnecessarily impairing the tagged sh's response to predators and to be able to immediately release the sh after surgery. Using anesthesia during sh surgeries may not always be the best course of action in every situation, and all potential effects on the sh must be considered (Damon-Randell et al. 2010; Lowerre-Barbieri et al. 2014). We believe the bene ts of reduced total handling time for sh tagged in situ versus sh tagged on the boat, which usually included a recovery period (mean ± SE: In situ = 9.30 ± 0.31 min, Boat = 72.2 ± 6.00 min), coupled with immediate release into protective habitat and the lack of pressure and temperature changes, explain the increased survivorship for in situ tagged sh. The increased survival of sh tagged in situ compared to sh tagged on the boat indicates that any negative effects due to the lack of anesthesia are minimal and do not outweigh the bene ts of in situ tagging.
There were several surprising results. Some of the variables that were originally believed to impact survivorship were either not in uential in the hazards models or had the opposite effect. For example, we hypothesized that additional handling time would increase the risk of a tagged sh having an event. The models, however, suggested an increase in 1 min of handling time slightly decreased the risk of hazard, though the con dence intervals neared or crossed 1 in all cases, indicating little to no effect. Handling time may not have been very in uential in survival, but it is still surprising that increased time may have reduced risk, especially as minimizing handling of sh has been reported as one of the most important considerations in implantation surgeries (Rub et al. 2014). In the case of boat-tagged sh, where handling time had a marginal but signi cant effect, an increase in handling time could have resulted in better recovery from the anesthesia. For in situ tagging, handling time was not signi cant. Handling time was not included in the initial hazards model because of correlation with tagging method, but we believe that 1) the reduced handling time during in situ tagging contributed to the high survivorship and 2) decreased risk with increased handling time for boat tagged sh was due to the sh recovering from anesthesia.
Fish size was not a signi cant factor in the best tting models. Fish tagged in this study varied between 32-107 cm TL, a range large enough that differences in survivorship because of size were expected. We hypothesized that smaller sh would be at a higher risk of an event because they might not have the same energy resources to aid in recovery as larger sh or they may be more susceptible to predation. However, this was not supported by the models. "Fish size" was included in the best tting model for in situ tagged sh at 6 days after tagging but was not signi cant, and a large spread in the hazard ratio con dence interval (0.77-20) represented high uncertainty in the hazard ratio estimate. While there are likely species-speci c differences (e.g, body type, stress tolerance) that in uence survival of different sized sh, in our study sh size did not signi cantly affect survival. Also, sh less than 20 cm FL have been shown to have high survival from internal tagging (Klinard et al. 2018), suggesting that surgery and implantation of appropriately sized tags for the body size of the sh may not impair survival, regardless of sh size.
Fish family was also not a signi cant covariate, although differences in tolerance to barotrauma and handling were seen among species. The high number of species in this study (which originally led to overparameterization in models) and the unevenness of species between tagging methods resulted in grouping species by family. There are physiological differences between species of the same family, and grouping them by family may have masked these differences and may help explain why family did not affect survivorship. Although an in-depth look at the effect of tagging method on species was not possible in this study, a planned, paired study with fewer species may nd that sh species in uences survival probability. For any tagging project, factors such as body type, susceptibility to barotrauma, and stress tolerance should be considered when deciding on the appropriate tagging method.
Experience of the surgeon has been shown to in uence tag retention and survivorship (Cooke et al. 2003; Deter et al 2010), but this was not seen in our study. In fact, surgeon experience was not a covariate in the top three parameterizations of any hazards model. Consistent training among surgeons may have reduced the effect of surgeon experience enough so that other variables had larger in uences. Before the start of a tagging project, surgeons practiced the tagging procedure with guidance from an experienced team member, and initial training was provided by a professional plastic surgeon (D. Hawtof, personal communication). Additionally, less experienced surgeons were always paired with an experienced surgeon during the tagging procedure to provide guidance during both boat and in situ tagging. The training and oversight throughout the tagging procedure may have helped to reduce the effect of surgeon experience on survivorship.
Hazards models were analyzed at both 4 and 6 days after tagging because there was discrepancy in the literature about how long after surgery to examine survival related to the tagging procedure. Results between these two time periods were similar (e.g., hazard ratios for tagging method in the initial model differed by only 0.015), but we felt it was informative to present both cases. Although differences between the results at both time periods were minimal, they are still interesting to note because they show how the dynamics of sh survival may subtly shift over time.
This study was not originally designed to rigorously examine all factors affecting boat-and in situ tagged sh. Therefore, some variables may not have proven as in uential as they would in a more systematic study. Depth was expected to have more of an in uence on hazard because detrimental effects of barotrauma have been shown to increase with depth of capture (Hannah and Matteson 2007). In this study, depth may not have been the best representation of the severity of barotrauma. Boat-tagged sh were generally caught using hook and line and were not necessarily caught from the sea oor, which was the depth recorded, and furthermore, mean depth of capture of boat-tagged sh was signi cantly shallower compared to in situ tagged sh (Wilcoxon two-sample rank test, p < 0.05). Fish that were near the surface or mid-water column when caught on hook and line would have a greater depth recorded than where they were physically caught in the water column, meaning that they may have exhibited fewer signs of barotrauma than if they had actually been caught at the depth recorded. Fish tagged in situ were caught via baited traps on the sea oor, so the depth of capture was the depth of the trap and the actual depth of the sh when it entered the trap. Additionally, after the start of in situ tagging, this method was preferentially chosen over tagging on the boat at deep sites and only a few sh were captured, tagged, and released from the boat in deep water (> 30 m). The method of tagging was based on what was deemed best for survival of the sh and logistically made sense for each tagging event. If boat tagging data had included more sh captured from deep water, resulting in no discrepancy in mean depth per tagging method, we hypothesize depth would have had a larger effect on survival. Additionally, comparing survivorship between tagging methods where boat-tagged sh were either released at the surface or brought back down to depth for release (the method used for all boat-tagged sh in this study) would further explore the effects of depth, barotrauma, and release method.
Fight time, location of hook placement on sh, whether the sh exhibited signs of barotrauma, whether a sh with barotrauma was vented before its return to depth, and overall health of the sh at time of release were not often recorded and thus were not included as covariates in the hazards models. All these factors could have affected the body condition and level of stress of the sh. Further testing of survival between sh tagged on the boat and sh tagged in situ would bene t from a planned, paired experiment to better determine how some of these other variables affect hazards to survival of internally tagged sh. This study also demonstrated some of the limitations of a posteriori-designed study. Unfortunately, additional tagged sh were excluded because of missing data, particularly during the initial stages of conducting acoustic telemetry research when it was not known what supplementary data would be bene cial to evaluate survivorship.
Despite these caveats, we believe that this historic dataset using over 100 internally tagged sh representing 14 different species surgically tagged by 11 different scientists was robust enough to demonstrate that tagging method clearly had the largest effect on sh survival. Telemetry studies are expensive to conduct, both in terms of effort and money, and often outcomes are not known until months later when the data can be downloaded. Therefore, it is critical to have a better understanding of how tagging method in uences sh survivorship. By examining the effects of tagging methods, we can inform future telemetry studies, thereby increasing the usefulness of telemetry data.
This study has shown that there are clear bene ts to tagging sh in situ. However, we understand that this method may not always be practical. In situ tagging can increase the amount of effort spent to tag each sh because an entire dive team is needed to tag one sh and there is limited time when tagging at deeper depths. Some target species may be located outside recreational diving limits, which can increase effort and cost for specialized training and dive gear. Studies that rely on anglers to supply the tagging subjects or focus on species found in very shallow water may not nd in situ tagging feasible or worthwhile. In this study we used baited traps to capture sh in situ, but other methods should be considered if the species of interest is unlikely to be captured in this manner.

Conclusions
In this study, sh internally tagged in situ with acoustic coded tags were ~ 75% less likely to suffer an "event" (mortality, tag loss, or emigration) shortly after tagging compared to sh brought to the surface and tagged on a boat. Other factors had little to no in uence on survival probability. Therefore, when designing a telemetry study, serious consideration should be given to the tagging method because it could play a critical role in the overall health of sh and success of the study. Implanting internal acoustic tags in situ is not a practical method for every species of interest and for every environment, but given the increased sh survivorship demonstrated here, we strongly suggest it be considered as the preferred tagging methodology for reef sh species where applicable.    Forest plots showing covariate hazard ratios at 4 days after tagging for the a) full model and b) besttting Cox proportional hazard model. Hazard ratios and 95% con dence intervals for each covariate are shown with p-values listed on the right-hand side of the plots with asterisks indicating signi cance.
Hazard ratio values less than 1 (represented by the dotted line) indicate a reduction in risk of an event while values greater than 1 indicate increased risk of an event. Covariates with hazard ratios close to 1 have a marginal effect on hazard and con dence intervals crossing 1 indicate no signi cant effect.