This item has the following downloads:
1 RESOLVING UNCERTAINTIES IN GULF STURGEON MORTALITY AND MOVEMENT RATES By MERRILL B. RUDD A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2013
2 2013 Merrill B. Rudd
3 To Mom, Dad, and Nate
4 ACKNOWLEDGMENTS I would like to thank my advisor, Dr. Rob Ahrens, and essentially co advisor, Dr. Bill Pine, for their teaching, mentorship, and encouragement throughout my graduate program. Thank you to my other committee member, Dr. Stephania Bolden, for her guidance through Gulf sturgeon management needs. I send a gigantic thank you to all of the Gulf sturgeon researchers from across the Gulf coast who have so generously shared their data and sturgeon knowledge with me: Ivy Baremore and Drew Rosati (NMFS), Ken Sulak and Mike Randall (USGS), Dewayne Fox, Naeem Willett, and Kate Fleming (DSU), Frank Parauka, Adam Kaeser, and Glenn Constant (USFWS), and Mark Peterson and his research team (USM). Thank you to Nicholas Cole for providing incredible support, encouragement, and belly laughs for an amazing two years of work and life. I would not have put in the extra ho urs necessary to do a great job without you next to me. I would also like to thank all of the students, faculty, and staff in Fisheries and Aquatic Sciences, especially in the Dequine Building, who have made my experience here memorable and helped talk thr ough models and research questions. Finally, thank you to my parents, for always offering help even when there is nothing they can do, and my brother Nathan, my fellow scientist, moral compass, and the kindest and most motivated person I know.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 7 LIST OF FIGURES ................................ ................................ ................................ .......... 9 ABSTRACT ................................ ................................ ................................ ................... 11 CH APTER 1 GENERAL INTRODUCTION ................................ ................................ .................. 13 2 MOVEMENT RATES AND GULF STURGEON STOCK STRUCTURE IMPLICATIONS USING ACOUSTIC TELEMETRY ................................ ................ 17 Methods ................................ ................................ ................................ .................. 20 Acoustic Tagging Program ................................ ................................ ............... 20 Multi state Model ................................ ................................ .............................. 21 Data Structure ................................ ................................ ................................ .. 22 Creating the Transition Matrix ................................ ................................ .......... 2 3 Transitions by natal river ................................ ................................ ............ 25 Transitions by genetic area ................................ ................................ ........ 25 Generated Data ................................ ................................ ................................ 26 Model Structures ................................ ................................ .............................. 27 Fitting model structures to generated data ................................ ................. 27 Fitting model structures to real data ................................ ........................... 28 Results ................................ ................................ ................................ .................... 29 Simulation Test ................................ ................................ ................................ 29 Movement Rates Estimated from Real Data ................................ .................... 29 Discussion ................................ ................................ ................................ .............. 31 3 SPATIAL DYNAMICS OF GULF STURGEON SURVIVAL AND DETECTION PROBABILITY FROM REMOTE ACOUSTIC TELEMETRY ................................ ... 50 Methods ................................ ................................ ................................ .................. 52 Real Data Structure ................................ ................................ .......................... 52 Generated Data ................................ ................................ ................................ 53 Model Structures ................................ ................................ .............................. 55 Fitting model structures to generated data ................................ ................. 55 Fitting model structures to real data ................................ ........................... 56 Results ................................ ................................ ................................ .................... 56 Simulation Test ................................ ................................ ................................ 56 Parameters Estimated from Real Data ................................ ............................. 58 Discussion ................................ ................................ ................................ .............. 61
6 4 STOCK ASSESSMENT OF GULF STURGEON USING AGE STRUCTURED MARK RECAPTURE ANALYSIS ................................ ................................ ............ 83 Methods ................................ ................................ ................................ .................. 85 Historic Mark Recapture Database ................................ ................................ .. 85 Initial Age Assignment ................................ ................................ ...................... 86 Age Structured Mark Recapture Model ................................ ............................ 88 Results ................................ ................................ ................................ .................... 92 Data Export ................................ ................................ ................................ ...... 92 Parameter Estimation ................................ ................................ ....................... 93 Natural mortality ................................ ................................ ......................... 93 Abundance of age 2+ and 4+ individuals ................................ ................... 93 Age 1 Recruitment ................................ ................................ ..................... 94 Capture probability ................................ ................................ ..................... 96 Exploring bifurcation in abundance est imates ................................ ............ 96 Discussion ................................ ................................ ................................ .............. 97 5 CONCLUSIONS AND RECOMMENDATIONS ................................ ..................... 116 LI ST OF REFERENCES ................................ ................................ ............................. 122 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 130
7 LIST OF TABLES Table page 2 1 Number o f acoustic transmitters surgically implanted in Gulf sturgeon annually by location. ................................ ................................ ........................... 37 2 2 Number of individual Gulf sturgeon acoustically detected during each month of the sampling period, us ed to verify seasonal groupings. Months 9, 10, 11 (September, October, November) and 3, 4, 5 (March, April, May) are peak migration periods. ................................ ................................ ............................... 38 2 3 Assumed parameter values for all tagged Gulf sturgeon used to generate detection histories to test whether the multi state model is unbiased. ................ 39 2 4 Proportion of estimation failure and confidence intervals not including true parameter values for fidelity and emigration rates. ................................ ............. 40 2 5 AICc table comparing models of Gulf sturgeon population dynamics for river specific movements. ................................ ................................ ........................... 41 2 6 A ICc table comparing models of Gulf sturgeon population dynamics for genetic area specific movements. ................................ ................................ ...... 42 2 7 Detection probabilities of Gulf sturgeon by river for in an d out migration periods with lower (LCL) and upper (UCL) confidence limits. ............................. 43 2 8 Detection probabilities of Gulf sturgeon by geographic areas based on genetic analysis with lower (LCL) and u pper (UCL) confidence limits. ............... 44 2 9 Transition probabilities of Gulf sturgeon movement between rivers; 95% confidence intervals in parentheses. ................................ ................................ .. 45 2 10 Transition probabilities of Gulf sturgeon between geographic areas based on genetic analysis; 95% confidence intervals in parentheses. ............................... 46 3 1 Proportion of estimation failure and confidence intervals not including true parameter values for state specific survival parameters and constant detection probability using the river specific spatial collapse method. ................ 69 3 2 Proportion of estimation failure and confidence intervals not including true parameter values for state specific survival parameters and constant detection probability using the genetic area spatial collapse method. ................ 70 3 3 Proportion of estimation failure and confidence intervals not including true parameter values for year specific and year and state specific survival rates. .. 71
8 3 4 AICc table comparing models of Gulf sturgeon population dynamics for real data under the river specific spatial collapse method. ................................ ........ 72 3 5 Gulf sturgeon instantaneous total survival es timates using the river specific spatial collapse method for river specific rates, with upper (UCL) and lower (LCL) 95% confidence limits. ................................ ................................ .............. 73 3 6 AICc table comparing models of Gulf sturgeon p opulation dynamics for real data under the genetic area spatial collapse method. ................................ ........ 74 3 7 Gulf sturgeon instantaneous total survival estimates using the genetic area specific spatial collapse me thod for natal area specific rates, with upper (UCL) and lower (LCL) 95% confidence limits. ................................ ................... 75 3 8 Gulf sturgeon instantaneous total survival estimates using the genetic area specific spatial collapse method for state specific rates, with upper (UCL) and lower (LCL) 95% confidence limits. ................................ ................................ .... 76 3 9 River specific detection probability estimates with upper (UCL) and lower (LCL) 95% con fidence limits. ................................ ................................ .............. 77 3 10 Detection probability estimates using the genetic area spatial collapse method from models assuming state specific and natal river specific survival, with upper (UCL) a nd lower (LCL) 95% confidence limits. ................................ 78 4 1 Number of initial marks and recaptures of Gulf sturgeon with internal and external passive tags in each river from 1977 2012. ................................ ........ 104 4 2 Total number of Gulf sturgeon and percentage of Gulf sturgeon tagged in each river that emigrated outside of the river in which they were tagged. ........ 105 4 3 Median estimates of Gulf sturgeon natural mortality from the MCMC posterior distribution with constructed upper (UCL) and lower (LCL) 95% confidence intervals. ................................ ................................ ................................ ........... 106
9 LIST OF FIGURES Figure page 2 1 Map of Gulf of Mexico and adjacent rivers with VEMCO VR2W acoustic receivers gating their entrances ................................ ................................ ......... 47 2 2 Example capture histories of Gulf sturgeon acoustically detected at river mouths, collapsed into discrete time interval s ................................ ................... 48 2 3 Box and whisker plots of the distribution of mean movement rate estimates representative for Gulf sturgeon from 1,000 iterations of generated data .......... 49 3 1 Box and whisker plots of the distribution of mean river specific survival and detection probability estimates representative for Gulf sturgeon from 1,000 iterations of generated data ................................ ................................ ............... 79 3 2 Box and whisker plot of the distribution of mean genetic area specific riverine and marine survival rates using the high resolution spatial collapse method applied to 1,000 iterations of ge nerated data ................................ .................... 80 3 3 Distribution of mean total year specific survival and year and state specific survival estimates from 1,000 iterations of generated data ............................... 81 4 1 The number of initial marks and recaptures of Gulf sturgeon at age over time. Numbers are expressed by bubb le size ................................ .......................... 107 4 2 Von Bertalanffy growth curve fit to limited age at length data shows a high variation in ages as length increases. ................................ ............................... 108 4 3 Abundance estimates in numbers for age 2+ Gulf sturgeon populat ions over time ................................ ................................ ................................ .................. 109 4 4 Maximum likelihood esti mates of abundance trajectories of Apalachicola River Gulf sturgeon compared with MCMC estimates ................................ ..... 110 4 5 Numbers of age 1 Gulf sturgeon recruits over time ................................ .......... 111 4 6 C apture probability of Gulf sturgeon over time ................................ ................. 112 4 7 Divergence in maximum likelihood parameter estimates between sample i terates of age assignment for the Apalachicola River Gulf sturgeon population ................................ ................................ ................................ ........ 113 4 8 Comparison of Poisson and negative binomial probability distributions for the number of predicted Gulf sturgeon recaptures in the year 2004 ...................... 114
10 4 9 Suwannee River Gulf sturgeon historic mark recapture data and population parameters estimated from CJS mark recapture model ................................ .. 115
11 Abstract of Thesis Presented to the Graduate School of the Universit y of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science RESOLVING UNCERTAINTIES IN GULF STURGEON MORTALITY AND MOVEMENT RATES By Merrill B. Rudd August 2013 Chair: Robert Ahrens Major: Fisheries and Aq uatic Sciences This study utilized the data collected from a standardized acoustic telemetry tagging program and historic mark recapture database t o resolve uncertainties in Gulf sturgeon mortality and movement rates at spatial scales relevant to manag em ent. Using multi state models with river mouth acoustic receiver detections over a two year period, I estimated detection probabilities greater than 0.50 across all rivers and time periods River and marine survival rates were similar in all regions except the western Gulf, where riverine survival was lower Gulf sturgeon exhibited non random fidelity to their natal rivers, and were more likely to enter their natal river or genetically related riverine population High fidelity rates to rivers or geneti call y distinct units combined with differential mortality rates spatially provides support for considering Gulf sturgeon stocks as distinct population segments. This delineation could be important if population segments show divergent evidence of recovery. U sing estimates of mortality from the telemetry study and a historic mark recapture database, I updated the Gulf sturgeon stock assessment using an a ge structured mark recapture (ASMR) model. I fit initial marking and recapture data from the Suwannee, Apala chicola, and Choctawhatchee Rivers incorporating uncertainty in
12 initial age assignment, to estimate a time series of abundance, recruitment, and survival in each riverine population Assuming a negative binomial probability distribution, model results wer e possibly divergent depending on pulsed or constant effort sampling designs. These results demonstrate important tradeoffs between monitoring time, size of the marked population, and capture probability, which are necessary to consider in managing a prote cted species.
13 CHAPTER 1 GENERAL INTRODUCTION Gulf sturgeon ( Acipenser oxyrinchus desotoi ) was listed as threatened under the Endangered Species Act (ESA) in 1991. Gulf sturgeon are anadromous fish living in estuarine and marine habitat during the winte r and migrating annually to river ine habitats for spawning and thermal refuge during the summer (Wooley and Crateau 1985) Currently, Gulf sturgeon utilize riverine environments adjacent to the Gulf of Mexico from Louisiana to Florida. Individuals have very high fidelity to their natal rivers, but occasionally emigrate to other river drainages after population mixing in centralized overwintering locations ( Stabile et al. 1996; Parauka et al. 2011 ) As a species listed under the ESA, Gulf sturgeon are currently managed as a single population and is not likely recovering at a rate to be delisted by the goal year 2023 (Flowers 2008) Life history characteristics such as spawning periodicity, slow growth, and specialized spawning habitats make Gulf sturgeon populations highly vulnerable to anthropogenic thr eat and lead to sl ow recovery. Mature f emales require more than one year between spawning events, resulting in only a fraction of the mature adult population spawning each year ( Huff 1975; F ox et al. 2000; Fox et al. 2002 ) Within t heir range, some rivers with Gulf sturgeon populations have limited access to upstream habitat due to obstructions such as dams (Pearl and Apalachicola Rivers), preventing individuals from reaching upstream spawning grounds with optimal substrate and flow rates (Fox et al. 2000) Gulf sturgeon are at risk for incidental bycatch from trawl and gill net fisheries, in cluding commercial shrimp fisheries in federal waters (NOAA et al. 2012 ) and a gar fishery using entanglement gear in southeast Louisiana ( USFWS and NMFS 2009) Point and non point discharges, coastal dredging, climate change, inadequacy of
14 regulatory mechanisms including limited enforcement, hurricanes, boat collisions, red tide, and risk of hybridization from escap ed sturgeon aquaculture operations are all listed as threats limiting the recovery of Gulf sturgeon ( USFWS and NMFS 2009) Since the Gulf sturgeon fishery was c losed and the species gained protected status, natural mortality ( M) was often considered equal to total mortality ( Z ) Total mortality is commonly estimated from tagging studies, as was conducted for Gulf sturgeon by Zehfuss et al. (1999) Sulak and Clugston (1999) and Berg (2004) Natural mortality can be derived from population dynamics and life history characteristics such as longevity (Hewitt and Hoenig 2005) temperature (Pauly 1980) and von Bertalanffy growth parameters (Gulland 1 983; Jensen 1996) Pine et al. (2001) estimated natural mortality from the Gulland method based on the von Bertalanffy shape parameter k, relating asymptotic length L to mean length at capture and vulnerability. Flowers (2008) used an alternative method to the Jensen method of M=1.5k instead assuming M=k (C. Walters, University of British Columbia, personal com munication). Gulf sturgeon n atural mortality estimates based on life history characteristics were consistently lower than estimates from mark recapture tagging programs. This difference indicated some anthropogenic mortality is impacting Gulf sturgeon that may not be consistent across populations Estimates of mortality from tagging programs were laden with high uncertainty due to low capture probabili ty and non cohesive tagging programs between river drainages ( Pine and Martell 2009) Previous tagging programs that estimated mortality relied on the physical recapture of external and PIT tagged fish, with capture probability ranging between 0.10 and 0.15. Low capture probabilities are common in aquatic
15 studies (Pine et al. 2003) resulting in high uncertainty in m ortality estimates due to inadequate inference in the fate of the un captured fish. When a fish is not captured over a series of time periods, the fish could be dead, not present in the sampling area, or was present but not captured. Gulf sturgeon riverine population mixing limits our understanding of Gulf sturgeon stock structure, anthropogenic and natural threats affecting Gulf sturgeon survival, and our ability to homogenize capture probability in rivers. The 2009 Gulf sturgeon stock assessment identifie d the uncertainty in mortality estimates as high priority in order to verify the perceived differences in recovery cross the geographic range in order to manage the species across its large spatial scale (Pine and Martell 2009) Specific m anagement recommendations in the 2009 stock assessment focused on improving mark recapture programs across the Gulf of Mexico and increasing data availability to all research groups. Implementing temporally and spatially consi stent monitoring programs to increase capture probability of tagged Gulf sturgeon would provide better estimates of stock assessment parameters (Pine and Martell 2009). Additionally, the 2009 stock assessment identified the need for a standardized, central ized database to archive historical data collection and update coordinated ongoing tagging efforts (Pine and Martell 2009). Since 2009, NMFS has initiated a standardized acoustic telemetry tagging program, creating acoustic receiver arrays at Gulf river mo uths and using standardized acoustic transmitters and PIT tags. NMFS has also created a centralized database for tagging information and assembled historic mark recapture data with information on all PIT tags, T bar, and acoustic transmitters associated w ith each individual.
16 The overall objective of my thesis research was to develop a framework for estimating mortality and movement patterns in Gulf sturgeon and use this information to develop new, and update existing, modeling frameworks for Gulf sturgeo n stock assessments to inform management decisions for this species. Chapters 2 and 3 utilize data from standardized acoustic telemetry tagging program to estimate demographic parameters for Gulf sturgeon, including survival and movement, while Chapter 4 u tilizes an ongoing mark recapture database to complete an age structured stock assessment for this species. In Chapter 2, I describe and evaluate a mark recapture modeling framework to estimate movement rates between spatial regions of the Gulf of Mexico including a simulation model to assess uncertainty in parameter estimates. In Chapter 3, I fit this model from Chapter 2 to data from the first 2 years (of a 5 year study) to estimate survival and detection probabilities for Gulf sturgeon at meaningful spa tial scales. For example, we estimate area specific survival to identify if there are differences in survival rates between Gulf sturgeon populations in the eastern and western Gulf. I also provide guidance on how additional years of data will improve infe rence. In Chapter 4, I incorporated uncertainty in initial age assignment to the age structured mark recapture model and present stock assessment results for three rivers with Gulf sturgeon spawning populations. I then combined the findings from these anal yses to make recommendations for Gulf sturgeon management and future monitoring programs in Chapter 5
17 CHAPTER 2 MOVEMENT RATES AND GULF STURGEON STOCK STRUCTURE IMPLICATIONS USING ACOUSTIC TELEMETRY Identification of stock structure is fundamental to f isheries management (Begg and Waldman 1999) Understanding how stocks are spatially defined is crit ical to conservation efforts, providing strategic utility in management and resource allocation ( Hilborn 1990; Ebbin 1996; Begg and Waldman 1999 ) For rare or protected species, management plans often specify that actions should be taken to reduce the risk of extinction or promote population recovery to certain levels. If differential threats to achieving th ese goals exist within their range, i t may be more appropriate to manage population units separately. Hilbor n and Walters (1992) defined stock as groups of fish large enough to be essentially self reproducing, with members of each group having similar life history characteristics. This definition encompasses a wide range of ideologies previously used to define the stock conc very low exchange rates (<1 per generation) are necessary to maintain genetic differentiation and distinct groups (Coyle 1998) From a taxonomic perspective, this definition may be adequate. In terms o f management and conservation of a species, greater attention to the potential for managing different components of the population for different management objectives (such as harvests) or assessing threats to a species (for rare species or threatened popu lations) is key. Biologists often adhere to the populations may adapt separately to their respective environments, despite relatively high rates of interpopulation excha nge (Coyle 1998)
18 Traditionally, genetic analysis and morphological characteristics, amongst other methods, have been used to produce detailed information on population structure identity based on evolutionary time scales of development (Carr et al. 1995 ; Begg et al. 1997 ) In the long term, management should focus on conserving biodiversity of exploited fish populations (Carvalho and Hauser 1994) In the case that genetically distinct pop ulations of the same species were divergently overfished, management objectives would need to utilize the distinct stock concept. However, those methods are limited in their ability to identify how populations respond to dis turbance over shorter time perio ds. In this way, the dynamic stock concept, assessing behavior and life history of individuals of a species over less than one generation, is likely applicable in a management context. Defining stocks using alternate methods, including mark recapture and population parameters may be especially important for species that may move between areas with different threats at different life stages (Ihssen et al. 1981 ; Carvalho and Hauser 1994 ) Mark recapture studies can inform how species temporally and spatially interact and mix, providing direct evidence of movement ( Skillman 1989; Coyle 1998;Begg and Waldman 1999 ) Mark recapture methods were used to identify stock structure f or lake whitefish ( Coregonus clupeaformis; Casselman et al. 1981) northwest Atlantic herring ( Clupea harengus; Stobo et al. 1975) and Gulf of Mexico king mackerel ( Scomberomorus cavalla; Johnson et al. 1994) reflecting homing fidelity behavior in adults and movement from spawning sites by juveniles (Begg et al. 1997; Coyle 1998) Stock structure has also been identified using population dynamic parameters ( Cushing 1968; Ihssen et al. 1981; Ebbin 1996; Coyle 1998 ) Th ese parameters are particularly
19 useful when the main objective of stock identification is for management purposes (Ihssen et al. 1981) as this type of information is necessary for developing eff ective management plans. For example, Brown et al. (1987) showed that yellowtail flounder ( Limanda ferruginea ) stocks reacted independently to exploitation despite a 10% exchange rate. With the short term goals of perpetuati ng economic benefit and long term goals of sustaining biodiversity, fisheries management often approaches stock definition at some point along a spectrum between the distinct (genetic) and dynamic (environmental) stock concepts (Carvalho and Hauser 1994) Mark recapture methods for identifying animal movement patterns and stock structure require that the sample of marked animals is represent ative of the population as a who le. In turn, recaptures of these marked animals mu st be representative of the marked population (Begg and Waldman 1999) In many mark recapture studi es, a key challenge is recapturing a high proportion of the previously marked individuals from which to estimate parameters of interest (Pine et al. 2003) In many cases, particularly with aquatic species, capture probability is low ( Pine et al. 2003; Lauretta et al. 2013 ) Traditional studies where recaptures are based on the animals being physically captured often estimate capture probabilities between 0.05 and 0.15. In telemetry studies, a detection can be interpreted as of a tagged fish (Hightower et al. 2001) whe re the detection confirms the (1) identity of the fish, (2) the location, and (3) the date /time stamp of the observation. This virtual recapture approach can lead to much higher capture probabilities than traditional tagging programs (Ihssen et al. 1981) Higher capture probability expands the utility of a mark recapture program by allowing
20 for direct assessment of tagged animal movement patterns and survival ( Hightower et al. 2001; Heupel et al. 2006 ) The objectives of this chapter were to (1) develop and evaluate a modeling framework to estimate movement and survival parameters for Gulf sturgeon from acoustic telemetry based on tag life, number of tags deployed, and detection rate and (2) using the framework established in objective (1) fit data collected by research cooperators for Gulf sturgeon to estimate river fidelity and emigration rates. My methods provide an example for temporally and spatially collapsing continuous acoustic telemetry data into discrete intervals to estimate objective parameters without bias. Methods Acoustic Tagging Program In 2010, NMFS initiated a standardized acoustic telemetry tagging program to resolve uncertainties in n atural mortality and movement rates that were identified in the 2009 Gulf sturgeon stock assessment as high priority (Pine and Martell 2009) VEMCO VR2W (Vemco Amirix Systems, Halifax, Nova Scotia) acoustic recei vers were deployed during late summer 2010 at the mouths of nine rivers adjacent to the Gulf of Mexico known to support Gulf sturgeon spawning populations, including the Pearl, Pascagoula, Escambia, Blackwater, Yellow, Choctawhatchee, Apalachicola, Ochlock onee, and the Suwannee Rivers (Figure 2 1). In some rivers with braided mouths, receivers were placed in the main river mouth and a few nearby small distributaries considered in the same drainage. All Gulf sturgeon in this study were initially captured dur ing fall out migrations of 2010 and 2011 by cooperating state, federal, and academic scientists using drift or anchored gill nets. VEMCO V16 acoustic transmitters (6H ) were inserted surgically into the gastric cavity of captured individuals considered to b e adults (total
21 length > 1350 mm ) following standardized surgical procedures (NMFS and USFWS 2010) Acoustic transmitters were surgically i mplanted internally to reduce tag loss (Bridger and Booth 2003; Welch et al. 2007) Transmitters were uniquely coded a nd emitted a signal every 90 seconds with an operating life of about six years. I assumed a comm on starting date for all acoustic detections as June 2010 because individual receivers were deployed at different starting times. My study only includes a portion of the actual data that will be collected as part of this project as tags will be deployed th rough 2013 and receivers will remain in place through 2015 Specific sampling procedures and initial data associated with the standardized acoustic telemetry tagging program are detailed in NMFS reports (Baremore and Rosati 2011; 2012) The sample size of Gulf sturgeon with aco ustic transmitters deployed in each river varied from 1 to 55 per year, resulting in a total between 12 and 105 in each river over the two year period of data I used (Table 2 1). In general, the number of transmitters deployed in the western Gulf (Pearl an d Pascagoula Rivers) was lower than the number deployed in the central or eastern Gulf. Due to the placement of receivers at the entrances to rivers, most Gulf sturgeon detections occurred during the spring period of migration into the rivers (March May) and fall outmigration from the rivers to the sea (September November) (Table 2 2). Multi state Model To estimate area specific movement rates for Gulf sturgeon using the telemetry based mark recapture data, I used a multi state model in Program MARK. The multi state model estimates three parameters: capture probability, apparent survival, and transition probability. Capture probability is the likelihood an individual is physically recaptured or acoustically detected (detection probability) during each time interval.
22 Apparent survival is the joint probability of surviving to the next time step (true survival) and the probability that the animal is not present in the sampling area during the current time step (temporary emigration). Transition probability is the likelihood that an animal geographic (i.e. which river the fish enters during in migration ) but can also be p hysiological or behavioral (e.g. transitio n between immature and mature). The multi state model is related to the cl assic Cormack Jolly Seber model, sharing the basic assumptions that (1) all individuals in the population (both marked and unmarked) have the same capture probability, (2) all marked individuals have the same probability of dying or permanently emigrating, (3) tags do not increase the probability of death, and (4) tags are not lost or overlooked. The multi state model works in a maximum likelihood framework using a multinomial function to estimate the likelihood of the observed capture histories given the model structure (White and Burnham 1999) A key utility of the multi state model is the ability to estimate area specific parameters. In cases when 100% of individuals are available to be captured or detected during each time pe riod, incorporating the transition probability between states absorbs the temporary emigration probability from apparent survival, and the multi state model estimates true survival. In my case, however, there is some possibility that an individual violates my assumptions about movement through their annual migration, so I conservatively assume my model estimates apparent survival. Data Structure I used knowledge about the biology of Gulf sturgeon to develop a framework for structuring data that would be a ble to estimate my objective parameters in the complementary multi state modeling framework. With virtual recaptures based on
23 telemetry relocations, a tagged Gulf sturgeon could be detected hundreds of times within a given day if the fish remained stationa ry within the detection range of the acoustic receiver. For the purposes of estimating movement, I chose to bin data at monthly time steps to inform the model time step This is based on knowledge of Gulf sturgeon life history where fish are thought to mov e into river systems from estuarine environments during early spring, remain in th e rivers during summer, and then move in the fall from the rivers into the estuarine and marine habitats (Wooley and Crateau 1985) Because of these movement patterns, I anticipated that detections on the receivers at the mouth of each river would be sparse except during periods of in migration or out migration between river and marine habitats. Because no data were added during these periods when Gulf sturgeon movement was not occurring, the lack of data would not add information to the likelihood probability Therefore I temporally collapsed the monthly detections into two seas ons per year, the out migration (September February) and the in migration (March August), to maximize information contributed from each time interval (Figure 2 2). Grouping the data this way satisfies the model assumption that each discrete time interv al is the same length. Creating the Transition Matrix Detections were denoted with an alphabet location code to describe the river drainage where the fish was detected during the time interval (nine possible river drainages). Due to the spatial setup of the sampling design the movement unadjusted represents the joint probability of a fish migrating into river A from the marine environment and moving out of river A after summer residence. To differentiate in migration return rates from out migration m ovement out of rivers, I added a marine state. All river drainage alphabet codes during out migration time periods were
24 M indicated movement into the estuary and marine environment (Figure 2 2). This method assumed a fish detected at the river mouth during the out migration moved into the estuary and stayed in the marine environment for the remainder of the fall and winter seasons. Under this structure, the return rate is represented as the probability of moving from the marine environmen t into a river ). The movement of Gulf sturgeon out of the rivers is represented as the probability of moving from a river into the marine environment ( ). I assumed 100% of fish left the rivers during out migration supported by Wooley and Crateau (1985) also facilitating model convergence by estimating fewer parameters. Estimating the transition parameter as state specific produces a transition matr ix to display movement probabilities between the marine and river environments. State specific movement rates should be interpreted as the probability of any fish entering each state, regardless of natal river or any previous location. Assuming population mixing at centralized overwintering marine habitats (Parauka et al. 2011) this model would assume individuals migra te into rivers randomly or based on preferred habitat or environmental factors. However, early studies of Gulf sturgeon migration support the hypothesis that individuals return to their natal river after winter population mixing (Wooley and Crateau 1985) ( Hestbeck et al. 1991; Brownie et al. 1993) considers second order Markovian transitions, whe re the state at time t+1 depends on a series of past locations prior to time t Although it is not possible for Program MARK to estimate the probability of movement based on a series of previous natal river and associated
25 Program MARK, I estimated the transition parameter dependent on state and natal river. Under this parameter structure, for example, I can es timate the probability of a fish from river A returning to river A separately from a fish from river B returning to river A However, movement between the marine environment and nine river drainages estimated separately for each focal river, represents 162 transition parameters alone. To limit the number of transition parameters and facilitate model convergence, I explored two approaches to collapse the data and reduce the number of terms estimated. Transitions by natal river In order to estimate river specific movement rates, I subset capture histories by natal river, so each of the nine subsets contained transmitters from only one river drainage. This method did not include all rivers in the same analysis. The number of states for the river specific m ethod varied by focal river. As an example, if no fish tagged in the Suwannee River emigrated during the time series, the analysis would only have two states (Suwannee River and marine). If fish tagged in the Suwannee River emigrated at any point to the Oc hlockonee and Apalachicola Rivers, the analysis would have four states (Suwannee, marine, plus two other rivers). Transitions by genetic area I collapsed the nine river drainages into four genetically distinct units by allele frequency supported by two g enetic analyses (Stabile et al. 1996) This allowed me to estimate ar ea specific parameters and included transmitters from all areas in t he same analysis. Genetic groupings included (1) West: Pearl and Pascagoula Rivers, (2) Escambia Bay: Escambia, Blackwater, and Yellow Rivers, (3) Choctawhatchee River, and (4) East: Apala chicola, Ochlockonee, and Suwannee Rivers. I assigned natal areas
26 to each capture history for the ability to estimate transition parameters as state and natal area specific. Generated Data Given limited sample size of tagged Gulf sturgeon and relativel y short term of the study, I conducted a simulation test t o verify the multi state model w ould estimate unbiased and precise parameter values under the given sampling design and model structure. I simulated tag detections using assumed survival, detection, fidelity, and emigration probabilities informed by previous studies ( Table 2 3; Huff 1975; Sulak and Clugston 1999; Fox et al. 2000; Berg 2004; Ross et al. 2009 ) The data generated extended over a two year period representing the initial two years of the five year NMFS standardized acoustic telemetry tagging program. Capture histories were generated first on a monthly time scale then collapsed into in and out migration seasons, mimicking my temporal collapse app lied to receiver detections described above. The simulation used Bernoulli trials consisting of three basic steps for each individual: (1) Hypothetical location which river is the fish near each month, based on its tagging location? (2) Survival given the next month? (3) Detection given the fish survives, is it detected? I conducted these simulations for both the river specific and genetic area specific methods of spatial data collapse. Under the river specific method, I tested three levels of tag sample size to include a range of the number of transmitters deployed : low (n=28 transmitters), medium (n=40 transmitters), and high (n=105 transmitters). These levels were relative to the number of tags biolo gists in each r egion were able to deploy over two years (Table 2 1). The genetic area specific method included various sample sizes of deployed transmitters from each genetic area in the same analysis.
27 Model Structures I tested several model structures to estimate fidelity and emigration rates without estimation failure. Under the river specific spatial collapse method, state dependent transitions inherently estimate non random, Markovian emigration because each analysis considers transmitters from one f ocal river at a time. Under the genetic area spatial collapse method, the transition parameter was considered either random (state dependent) or non random Markovian (state and natal river dependent). For both spatial collapse methods, I explored constant migration season, and state dependent detection probability as well as constant and state dependent survival rates. Models were built using all possible combinations and compared using Akaike Information Criterion corrected for small sample size (AICc). Fitting model structures to generated data I ran the respective model structures described above for each spatial collapse method using generated data. AICc should provide the most support for Markovian emigration with constant detection probability and area specific survival, since this was the model used to simulate the data. I conducted 1,000 iterations of data generation and their respective model runs for each spatial collapse method to understand the accuracy of model estimates. I calculated the n umber of model runs that resulted in estimation failure, defined as a parameter value that cannot be estimated given the data available. Estimation failure is often manifested as a lower confidence limit of 0 (or near 0) and an upper confidence limit of 1, lower and upper confidence intervals equal to the maximum likelihood estimate, or non number confidence limits (NA). In the river specific spatial collapse method, estimation of transitions that do not occur in the dataset would result in failure.
28 Althoug h these instances cannot be counted in the simulations, their existence is just something to consider. For example, if the Suwannee River is the focal river, estimating the probability a Gulf sturgeon would move between the Suwannee and Pearl River will re sult in failure because no fish were observed moving from the Suwannee to the Pe arl. I also measured model inaccuracy as the number of iterations where the true movement rate did not fall within the 95% confidence intervals. Precision of parameter estimate s was approximated using the range of mean parameter estimates from simulated iterations. Fitting model structures to real data Although the same models were run for each river and spatial collapse method, AICc values could not be directly compared betwe en these spatial scales and structures due to the way data were split and grouped. AICc measures the relative goodness of fit of a model (Akaike 1973) therefore two non identical datase ts cannot be directly compared using AICc. Each analysis under the river specific spatial collapse method utilized a completely unique set of transmitters divided by natal river. The genetic area spatial collapse method utilizes all transmitters in the sam e analysis. The genetic grouping further adjusts the data structure (individual rivers versus rivers grouped by genetic relatedness). Even if the same model has the lowest AICc for multiple rivers or spatial collapse methods, the movement rates would refer to spatially different occurrences. State dependent transitions are river specific under one collapse method (e.g. Marine Suwannee ) and genetic ar ea specific with the other (e.g. Marine East ). To simplify analysis, I chose only one parsimonious model for all rivers under the river specific collapse method, but considered the implications of different rivers having divergent model fits. I chose the most parsimonious model for the genetic area specific
29 collapse method independently of the river specific coll apse method, with the goal of estimating movement rates without estimation failure. Results Simulation Test The simulation test showed that on average, the multi state model estimated unbiased parameter values given the sampling design and my model struct ures (Figure 2 3). Precision of mean parameter estimates around the true parameter value increases with increased number of tagged Gulf sturgeon. Mean parameter estimates from an area with a low sample size of transmitters deployed had a much greater range around the median mean estimate than those from an area with a high sam ple size (Figure 2 3). When few Gulf sturgeon were tagged in a riverine environment a higher proportion of model runs resulted in estimation failure and inaccuracy (Table 2 4). Movem ent Rates Estimated from Real D ata Under the river specific spatial collapse method, I chose to analyze movements using the model assuming constant survival and migration season dependent detection probability. This model had the most AICc support for fou r of the ni ne river populations (Table 2 5 ). The model with constant survival and detection probability best fit the data from five of the nine riverine populations of Gulf sturgeon but in each of these cases the respective model considering migration sea son dependent detection probability was not significantly different (delta AICc < 3). For the four rivers with the most support for season dependent detection probability, the model assuming constant detection probability was significantly different (delta AICc > 3). This information complemented my interest in understanding how detection probability differed between the in migration and out migration, resulting in my model choice for estimating river specific rates. To
30 estimate movement rates between genet ic units, I chose the model that estimated state specific survival and detection probability with Markovian emigration, which received the most AICc support and meets Gulf sturgeon m anagement objectives (Table 2 6 ). Under both spatial collapse methods and across all states using real data, detection probability was estimated to be much higher than the conservative assumed detection probability utilized in data generation (Table 2 3 ; Table 2 7; Table 2 8 ). Therefore, I can expect a higher chance of accuracy than was observed in the simulation. River specific fidelity rates could be estimated without failure for all rivers except the Pearl, Pascagoula, and Suwanne e (Table 2 9 ). Tagged individuals from the latter three rivers never emigrated to any other riv er drainage within the two year period analyzed, resulting in estimation failure for the fidelity parameter. Estimation failure was likely because the model had difficulty estimating the parameter at its upper boundary. Gulf sturgeon had re latively high fi delity estimates for other riverine populations (Table 2 9 ). When Gulf sturgeon emigrated they generally moved into nearby rivers Gulf sturgeon movement rates were greater into rivers within areas of similar genetics compared to nearby rivers with popula tions that are not genetically (Table 2 9 ). For all movements that occurred during the two year period, their probability was able to be estimated without failure. Genetic area f idelity rates were higher than individual fidelity rates to rivers identified genetically (Table 2 10 ). Even though I pooled data from Gulf sturgeon tagged in the Pearl and Pascagoula Rivers in to a western Gulf
31 emigration estimated resulted in failures at the upper and lower boundaries because Gul f sturgeon did not move into other geographic areas during the two year period. Discussion The long term recovery objective identified in the 1995 Gulf sturgeon Recovery P geon in (USFWS 1995) Discrete management units could be def ined loosely based on river drainage or genetic units (Stabile et al. 1996) However, because the Gulf sturgeon is listed as a species throughout their range, only the species as a whole could be considered for delisting. Before even considering a potential delisting, a ssuming recovery criteria are met, delisting Gulf sturgeon by anything other than th e species would first require inct population The tagged Gulf sturgeon in this study expressed very high homing fidelity to the rivers where they were originally tagged. Gulf sturgeon had a relatively low probability of emigrating outside of their genetic area. Although population mixing is a function of distance my initial estimates suggest higher rates of movement between the eastern and central Gulf, with very low rates of movement in and out of the western Gulf. Even over the short time period analyzed, estimates of detection probability (0.46 0.95) were higher than estimates from PIT tag programs (0.10 0.15) (Pine and Martell 2009) enabl ing me to estimate river return rates with reasonable confidence intervals (range of 0.10 for genetic areas, 0.20 0.30 for rivers). However, given latitudinal similarity across the geographic range of Gulf sturgeon populations, homing fidelity is not enoug h to define Gulf sturgeon riverine populations or genetically distinct units as markedly
32 separate, important populations. More information would be needed to prove significance of treating Gulf sturgeon riverine populations differently through DPS criteria In this study, I assumed 100% movement out of rivers after the summer months and zero tag loss. The detection, survival, and movement rate parameters are potentially confounded (Schaub et al. 2004) With very low detection probability, it is difficult to infer whether un detected individuals died, remained in the marine environment over summer months, did not migrate out of the river after summer inhabitance, or moved through the river mouth out of detection range. Informed assumptions are useful in elimin ating the uncertainty around some of these possibilities, allowing for better inference in objective parameters. Field biologists on the Suwannee River regularly catch Gulf sturgeon at holding sites throughout the summer months, but individuals become spar ser into August and September (K. Sulak, USGS, personal communication). Similarly, fall out migration tagging efforts on the Apalachicola River and others sampled by USFWS are often deemed risky based on the fear of missing the majority of individuals movi ng out into the marine environment (F. Parauka, personal communication). If my assumption of 100% movement out of rivers is violated in reality, my estimates of Gulf sturgeon survival and detection probability would be biased An alternative to the assumpt ion of 100% movement out of rivers could be 100% detection, but this is much less realistic due to the branching nature of river mouths and limits of technology. Additional acoustic coverage with more receivers at the mouth of each river, combined with rec eivers upstream and in the marine environment, would likely increase detection rate and potentially more accurately confirm river and marine locations of Gulf
33 sturgeon. However, given the fixed array coupled with the life history of this anadromous fish, I assumed 100% movement out of rivers (Lebreton et al. 1992) Although this study assumes ze ro tag loss, a low portion of the tags are thought to be shed (D. Fox and N. Willett, Delaware State University, personal communication). With surgical implan tation, tag loss is possible due to bacterial infection at the wound (Daniel et al. 2009) or sutures dissolving before the incision is healed (Bridger and Booth 2003) Tag shedding and mortality are difficult to separate. The assumption of zero tag loss would lead to negative bias in survival rate estimates if the rate was in fact higher, further increasing uncertainty around all parameters (Arnason and Mills 1981) Uncertainty in Gulf sturgeon acoustic transmitter tag shedding could be determined thr ough an alternative application of the multi state model (Conn et al. 2004) All Gulf st urgeon surgically implanted with acoustic transmitters in this study are also tagged given known initial tags (i.e. acoustic and PIT tag, acoustic only, PIT only). Becau se Gulf sturgeon are long lived, a negatively biased survival rate estimate would impact my ability to assess how Gulf sturgeon recovery from mortality events and habitat changes. Another modeling issue was difficulty estimating movement rates at the uppe r and lower boundaries (0 and 1). Estimating fidelity rates to the Pearl, Pascagoula, and Suwannee Rivers resulted in failure, manifested as non number confidence intervals (NAs). Ad hoc observation of the data showed no individuals from these rivers emigr ated during the two year period. The estimation failure was not due to low sample size. Other riverine populations had lower sample siz es of tagged Gulf sturgeon than the Suwannee River (e.g. Ochlockonee River), but the fidelity rate was still estimation
34 w ithout failure. Furthermore, the fidelity rate estimation failure persists when the sample size of deployed transmitters in the Pearl and Pascagoula Rivers are pooled under the genetic area spatial collapse method. Estimation failure at the upper and lower boundaries could be resolved with adjustment of the link function used in Program MARK. This study utilized the default multinomial logistic regression (logit) link function for estimating transition parameters, as it is the only link function that constr ains transitions from the same state to sum to 1.0 (White and Burnham 1999) However, logit link functions are renowned for having difficulty estimating parameters at the boundaries. This issue was not immediately apparent from results of the simulation test because my assumed fid elity rate (0.80) was lower than observed rates. Although I have not currently worked through a method of estimation movement rates at the boundary with adequate certainty, it is clear for management purposes that fidelity rates in these areas could be ext remely high. The two spatial collapse methods represent variations in the presentation of emigration probabilities. An advantage of the river specific spatial collapse method is the ability to estimate, with adequate certainty, movements between river dr ainages that did occur during the time period. However, this method only provides estimates for the emigrations that occurred within the two year period. The maximum likelihood framework of analysis can only estimate the likelihood of an observation. For e xample, although the possibility of a Gulf sturgeon tagged in the Suwannee River moving into the Ochlockonee River is entirely logical given their close proximity, it was not observed during the two year monitoring period likely due to low sample size. The genetic unit spatial collapse method is a useful tool for estimating area specific survival rates, if Gulf
35 sturgeon movement is restricted geographically or between genetically similar e analysis, the model estimates transition probabilities between each (albeit low and with estimation failures). Especially due to the short time period of telemetry detections and limited sample size of transmitters monitored, movement rates need to be pr esented as probabilities, rather than number of individuals or presence/absence. The spatial collapse utilized in this study and associated movement rates of Gulf sturgeon between regions of the Gulf of Mexico are important for understanding the spatial di fferences in mortality and appropriately managing Gulf sturgeon populations. The results of this study have significant management implications for Gulf sturgeon under ESA as well as Natural Resource Damage Assessment (NRDA) legislation. Quantifying rive r return rates across the Gulf addresses goals for the NRDA investigation following the Deepwater Horizon Oil Spill (NOAA 2012 ) NRDA is responsible for conducting assessment of Gulf sturgeon in relation to oil inflicted areas. This study demonstrates that in the two years directly after the oil spill, Gulf sturgeon continued to exhibit high fidelity rates to both their genetically distinct regions and natal river drainages. The eastern and western areas of the Gulf exhibited the highest probability of site fidelity (0.98 and 1.00, respect ively). These findings would imply that area specific anthropogenic threats non randomly affect Gulf sturgeon populations due to the low movement rates between far away areas. Therefore, I would expect Gulf sturgeon exposed to area specific anthropogenic t hreats to be in greater danger than populations on the other side of the Gulf. Because this study utilizes a greater sample size of tagged Gulf sturgeon than any other published study of winter movement (e.g.
36 Ross et al. 2009; Parauka et al. 2011), these r esults provide greater inference for scientists to understand Gulf sturgeon population dynamics and estimate demographic parameters. I believe both NMFS and NRDA should utilize the framework developed, with the additional three years of data collected in t he acoustic telemetry monitoring program, to update the movement rates of Gulf sturgeon throughout the Gulf of Mexico.
37 Table 2 1. Number of acoustic transmitters surgically implanted in Gulf sturgeon annually by location. River 2010 2011 Pearl 11 1 Pascagoula 15 13 Escambia 30 14 Blackwater 31 20 Yellow 33 21 Choctawhatchee 55 50 Apalachicola 20 20 Ochlockonee 0 17 Suwannee 20 20 Total 215 176
38 Table 2 2. Number of individual Gulf sturge on acoustically detected during each month of the sampling period, used to verify seasonal groupings. Months 9, 10, 11 (September, October, November) and 3, 4, 5 (March, April, May) are peak migration periods. Month /Year # of Fish Detected 6/10 5 7/1 0 0 8/10 2 9/10 60 10/10 149 11/10 25 12/10 3 1/11 0 2/11 2 3/11 61 4/11 101 5/11 30 6/11 7 7/11 8 8/11 22 9/11 28 10/11 206 11/11 94 12/11 21 1/12 8 2/12 19 3/12 172 4/12 116 5/12 45 6/1 2 18
39 Table 2 3. Assume d parameter values for all tagged Gulf sturgeon used to generate detection histories to test whether the multi state model is unbiased. Parameter Assumed Value for Data Generation Annual marine survival 0.80 Annual river surviv al 0.90 Monthly detection probability ( Mar May and Sept Nov) 0.10 Monthly detection probability ( Dec Feb and June Aug) 0.02 Annual fidelity rate 0.80
40 Table 2 4. Proportion of estimation failure and confidence intervals not including true parameter values for fidelity and emigration rates. Note: O ut of 1,000 iterations of generated d ata and model runs b a sed on variable number of tagged Gulf sturgeon in each riverine population (upper portion ) and pooled number of tagged Gulf sturgeon in each genetically distinct unit (lower portion). Estimation Failure True Value Not I n CI River specific transitions Fidelity Emigration Fidelity Emigration Low sample size 0.21 0.08 0.16 0.62 Medium sample size 0.10 0.06 0.09 0.48 High sample s ize 0.01 0.03 0.05 0.16 Genetic Area Transitions Pooled Low 0.07 0.08 0.11 0.37 Pooled Medium 0.01 0.07 0.05 0.11 Pooled High 0.00 0.07 0.06 0.10 Pooled Highest 0.00 0.02 0.06 0.06
41 Table 2 5. AICc table comparing models of Gulf st urgeon population dynamics for river specific movements. Model Lowest delta AICc S(.)p(.)Psi(.) Suwannee, Apalachicola, Choctawhatchee, Yellow, Pascagoula, Pearl S(.)p(m)Psi(s) Ochlockonee, Blackwater, Escambia S(.)p(s)Psi(s) 0 S(s)p (.)Psi(s) 0 S(s)p(m)Psi(s) 0 S(s)p(s)Psi(s) 0 S(.)p(s*m)Psi(s) 0 S(s)p(s*m)Psi(s) 0
42 Table 2 6. AICc table comparing models of Gulf sturgeon population dynamics for genetic area specific movements. ancy across states, times, and groups. Model delta AIC c S(s)p(s)Psi(s*g) 0 S(s)p(m)Psi(s*g) 2.4 S(.)p(s)Psi(s*g) 3.4 S(.)p(m)Psi(s*g) 4.9 S(s)p(s*m)Psi(s*g) 11.0 S(.)p(.)Psi(s*g) 13.3 S(.)p(s*m)Psi(s*g) 14.2 S(.)p(m)Psi(s) 799.2 S(.)p(s)Psi(s) 805.4 S(.)p(.)Psi(s) 807.7 S(s)p(s)P si(s) 808.9 S(s)p(.)Psi(s) 810.4 S(.)p(s*m)Psi(s) 815.7 S(s)p(s*m)Psi(s) 819.4
43 Table 2 7 Detection probabilities of Gulf sturgeon by river for in and out migration periods with lower (LCL) and upper (UCL) confidence limits. River Migration Period Estimate LCL UCL Pearl Out 0.55 0.23 0.8 3 In 0.84 0.47 0.97 Pascagoula Out 1.00 1.00 1.00 In 1.00 1.00 1.00 Escambia Out 0.47 0.29 0.66 In 0.79 0.64 0.89 Blackwater Out 0.46 0.26 0.67 In 0.76 0.57 0.88 Yellow Out 0.48 0.32 0.65 In 0.57 0.47 0.67 Choctawhatchee Out 0.67 0.53 0.79 In 0.68 0.59 0.76 Apalachicola Out 0.86 0.47 0.98 In 0.78 0.58 0.90 Ochlockonee Out 0.51 0.23 0.76 In 1.00 1.00 1.00 Suwannee Out 0.95 0.70 0.99 In 0.93 0.79 0.98
44 Table 2 8. Detection probabilities of Gul f sturgeon by geographic areas based on genetic analysis with lower (LCL) and upper (UCL) confidence limits. Genetic Area State Estimate LCL UCL West 0.65 0.48 0.79 Escambia Bay 0.71 0.62 0.79 Choctawhatchee 0.71 0.61 0.79 East 0.8 5 0.74 0.92 Marine 0.62 0.54 0.69 Note: West = Pearl and Pascagoula Rivers, Escambia Bay = Escambia, Blackwater, and Yellow Rivers, Choctawhatchee is the Choctawhatchee River by itself, East = Apalachicola, Ochlockonee, and Suwannee Rivers, and Marine is the pooled detection probability during out migration from all geographic areas.
45 Table 2 9 Transition probabilities of Gulf sturgeon movement between rivers ; 95% confidence intervals in parentheses. Pearl Pascagoula Escambia Blackwater Yello w Choctawhatchee Apalachicola Ochlockonee Suwannee Pearl 1.00 x x x x x x x x (NA,NA) Pascagoula x 1.00 x 0.02 x x x x x (NA,NA) 0.00,0.13) Escambia x x 0.69 0.24 0.10 0.05 x x x (0.55,0.80) (0.14,0.38) (0.04,0.22) (0.02,0. 11) Blackwater x x 0.13 0.29 0.16 0.01 x x x (0.07,0.25) (0.18,0.43) (0.08,0.29) (0.00,0.06) Yellow x x 0.02 0.35 0.72 x x x x (0.00,0.12) (0.23,0.49) (0.58,0.83) Choctawhatchee x x 0.15 0.10 0.02 0.93 0.05 x x (0.08,0.28) (0.04 ,0.22) (0.00 ,0.13) (0.87,0.97) (0.01,0.18) Apalachicola x x x x x 0.01 0.75 x x (0.00,0.06) (0.59,0.86) Ochlockonee x x x x x x 0.18 0.56 x (0.09,0.32) (0.25,0.82) Suwannee x x x x x x 0.03 0.44 1.00 (0.00,0.16) (0.18,0 .75) (NA,NA) Note: Columns indicate natal river and rows denote destination. Estimates along the diagonal represent fidelity rates. An d text indicates one of the three alternate forms of estimation failure (both 95% confidence limits equal to maximum likelihood estimate, lower confidence limit is at or near 0 and upper confidence limit is 1.0, or non number (NA) confidence limits).
46 Tab le 2 10 Transition probabilities of Gulf sturgeon between geographic areas based on genetic analysis; 95% confidence intervals in parentheses West Escambia Bay Choctawhatchee East West 1.00 0.01 0.00 0.00 (0.00,1.00) (0.00,0.05) (0.00,1.00) (0.00,1 .00) Escambia Bay 0.00 0.90 0.06 0.00 (0.00,0.00) (0.84,0.94) (0.02,0.12) (0.00,0.00) Choctawhatchee 0.00 0.09 0.94 0.02 (0.00,1.00) (0.05,0.15) (0.87,0.97) (0.01,0.09) East 0.00 0.00 0.01 0.98 (0.00,0.00) (0.00,1.00) (0.00,0.05) (0.90,0.99) Note : Columns indicate natal area and rows denote destinations. Estimates along the diagonal represent fidelity rates Italicized text indicates one of the four forms of estimation failure (no individuals moved between the natal area and destination, both 95% confidence limits equal to maximum likelihood estimate, lower confidence limit is at or near 0 and upper confidence limit is 1.0, or non number (NA) confidence limits).
47 Figure 2 1. Map of Gulf of Mexico and adjacent rivers with VEMCO VR2W acoustic receiv ers gating their entrances. Rivers with known Gulf sturgeon populations are, from west to east: Pearl, Pascagoula, Escambia, Blackwater, Yellow, Choctawhatchee, Apalachicola, Ochlockonee, and Suwannee. State and federal waters are indicated by white contou r line. [Courtesy of Amanda Frick, NOAA].
48 A B Fall Winter Spring Summer S 0 S 0 C Out migration In Migration S S D Out migration In Migration M S E Ou t In Out In M S M A Figure 2 2. Example cap ture histories of Gulf sturgeon acoustically detected at river mouths, collapsed into discrete time intervals. A) 12 months. B) Four seasons. C) T wo migration seasons D) Two migration se asons with a marine sta te. E) Two migration seasons over two years with an example emigration Sept Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug S 0 0 0 0 0 S S 0 0 0 0
49 A B Figure 2 3. Box and whisker plots of the distribution of mean movement rate estimates representative for Gulf sturgeon from 1,000 iterations of generated data Prec ision of estimates is compared between low, medium, and high sample size of tagged Gulf sturgeon. A) F idelity rate. B) E migration probability to the most extreme data point no l onger than range multiplied by the interquartile range, and individual points represent outliers. Green and blue lines indicate the true values of Gulf sturgeon fidelity and emigration rates, respectively, used in data generation.
50 CHAPTER 3 SPATIAL DYNAMI CS OF GULF STURGEON SURVIVAL AND DETECTION PROBABILITY FROM REMOTE ACOUSTIC TELEMETRY Estimating demographic parameters, such as survival rate and po pulation for over time or between habitats (Nichols and Pollock 1983) Capture recapture methods produce unbiased estimates of demographic parameters by following a representative sample of marked individuals over time, removing bias prevalent in population abundance estimates from assumptions about marked animals (Lebreton et al. 1992; White and Burnham 1999; Pine et al. 2003 ) However, precision around demographic parameter estimates decreases with low capture probability and emigration from the study site (Zehfuss et al. 1999) When uncertainty around demographic parameter estimates is great, it is nearly impossible to distinguish significant population change from expected variatio n in parameter estimates within large confidence intervals. Survival rate in particular is a critical aspect of population assessment. Population decline and eventual extinction occurs when the mortality rate exceeds the birth rate. The classic open popul ation Cormack Jolly Seber (CJS) mark recapture modeling framework estimates unbiased survival rates, while assuming all individuals within the population have homogenous and relatively high capture probabilities throughout the year and any emigration is pe rmanent (Lebreton et al. 1992; Schaub et al. 2004) However, animal behaviors such as fidelity to breeding habitat and migration can lead to non random movements of marked individuals, leading to temporary emigrati on and heterogeneous capture probabilities throughout the year (Kendall et al. 1997) The magnitude of capture probability also depends on the system, species, and
51 sampling method. In cl osed populations and terrestrial systems, capture probability is often relatively high (Lebret on et al. 1992) It may be possible to recapture or relocate close to 100% of marked individuals, leading to high precision in survival rate estimates (Hightower et al. 2001) However i n open populations, common in aquatic studies, individual capture probability is likely very low (< 0.50) and heterogeneous by individual due to potential emigration (Pine et al. 2003) In aquatic systems where individual capture probability is usually low, acoustic telemetry is increasingly used to track animal mo vements (Heupel et al. 2006) Passive acoustic receivers can be used to detect marked fish in a capture resighting study to estimate survival. This method maximizes effort without the man power, contributing to cost efficiency (Grothues 2009) Curtain array designs entail receivers deployed in a line, or series of lines, so detection ranges overlap to ensure resightings are noted on at least one of the receivers. Curtain arrays can be deployed at possible habitats o ver a large spatial scale to identify entry and exit from a specified habitat rather than fine scale habitat use (Heupel et al. 2006) The use of acoustic telemetry to indicate a recapture studies drastically increases capture probabilities by removing the need for physical recaptures and applying standardized effort across possible utilized areas over a large spatial scale. The increased capture probability and spatial coverage associated with acoustic telemetry also increases the ability to estimate survival rates across a heterogeneous landscape. Species with annual migrations encounter areas with varying degrees of threa t over relatively short time scales, including differential predation rates, environmental effects, or anthropogenic impacts. The nature of open populations and
52 migratory species makes it difficult to study all individuals throughout their geographic range leading to high uncertainty in the threats to which they are subject throughout the year. Multi state models bridge together several sub populations into a larger super population. This method transforms several open populations into a closed population with possibilities for movement between sub populations. I utilized the NMFS standardized acoustic telemetry tagging program applied passive acoustic telemetry across a large spatial scale. Acoustic telemetry was used to identify presence in a particular river mouth during peak migrations to categorize broad scale, regional movements, as opposed to small scale habitat use. This Gulf wide survey was developed in the hopes that detection probability using passive acoustic telemetry would be greater than the capture probability associated with sampling relying on physical recaptures. I used these data collected by NMFS to (1) conduct a simulation test to verify the multi state model can estimate spatially and temporally explicit, unbiased, and precise paramete r values over short and long monitoring periods and (2) use current data to estimate survival and detection probabilities for Gulf sturgeon at meaningful spatiotemporal scales. Methods Real Data Structure Using the same methods described in Chapter 2, I f iltered acoustic detections into individual capture histories on a monthly time scale. I then temporally collapsed the capture histories into two seasons per year, the in migration (March August) and out migration (September M migration river mouth detections, assuming 100% of fish move out of the rivers and into the marine environment for the remainder of the fall and winter seasons, subject to a shared marine
53 surviva l rate (Figure 2 2). I explored river spe cific and genetic area spatial collapse methods to limit the number of transition parameters and facilitate model convergence. The river specific spatial collapse method subset capture histories by natal river, so each of the nine subsets contained transmi tters from only one river drainage. The number of states depended on the number of states to which fish from the focal river emigrated during the time period. The genetic area spatial collapse method collapsed the nine river drainages into four areas based on genetic analyses (Stabile et al. 1996) The genetic area method allowed me to estimate area spe cific survival and detection probability and included transmitters from all areas in the same analysis. Genetic groups included (1) West: Pearl and Pascagoula Rivers, (2) Escambia Bay: Escambia, Blackwater, and Yellow Rivers, (3) Choctawhatchee River, and (4) East: Apalachicola, Ochlockonee, and Suwannee Rivers. Generated Data I repeated the simulation test described in the Methods section of Chapter 2 to quantify expected uncertainty and bias surrounding survival and detection probability parameters given limited sample size of deployed transmitters and monitoring time. I generated monthly capture histories based on assumed rates of Gulf sturgeon survival, detection probability, and movement rates (Table 2 3). I collapsed the monthly capture histories into two seasons per year (in migration and out migration), the same method applied to real acoustic telemetry data All out migration river mouth detections were assigned the marine state. I spatially collapsed data using the river specific method and genetic area method to facilitate model convergence. Data was simulated over two years and five years under each spatial collapse method to compare levels of certainty with increased monitoring time.
54 In terms of my objective estimates of area specific survival, the river specific spatial collapse method had the potential to estimate natal river survival (shared survival rate for all fish from each river) and state specific survival (separate survival rates for the focal river, marine environment, and rivers to wh ich fish emigrated). This method provides very high spatial resolution of survival rates, but may be limited by low sample size of deployed transmitters for some regions. State specific survival for the genetic area spatial collapse method assumes one shar ed marine survival rate for all Gulf sturgeon given high mixing in the Gulf of Mexico during winter months ( Ross et al. 2009; Parauka et al. 2011 ) and estimates separate riverine survival rates for each genetic un it I did not refine threats in the marine environment that may only affect certain populations of individuals. For example, shrimp trawling in the western and central Gulf may only affect fish traveling from western and central rivers to the shared, centr al sites for marine residenc e (NOAA et al. 2012) Natal river dependent survival is a more abstract concept, as survival is dependent not on a physical location, but an indivi specific survival under the genetic area collapse method estimates separated fish into genetic units and then estimates one annual survival rate for each genetic group. Marine and river survival rates are not estimated separat ely. Natal river dependent survival assumes high area fidelity, where fish are not moving between genetic areas and in turn subject to region specific threats. I tested an alternate data structure applied to the genetic area spatial collapse method to ac count for genetic area specific marine survival as well as genetic area river survival. The high resolution genetic area spatial collapse method assigned different X
55 d a fish detected entering the Suwannee River, XS XA I applied all three spatial collaps e methods (river specific, genetic areas, and high resolution genetic area) to generated data to assess whether we could expect unbiased results from the multi state model. Model Structures In order to estimate area specific survival and detection probab ilities, I used the multi state model in Program MARK (described in Chapter 2). In addition to estimating area specific parameters, I included parameter structures with a temporal component to identify significant population change from short term anthropo genic or natural impacts. Model structures in this analysis considered survival rates as state dependent, year dependent, natal river dependent, dependent on both state and year, and constant over state, year, and natal river. I considered detection probab ility as migration season dependent, state dependent, dependent on both state and migration season, and constant over state and season. I assessed all combinations of survival and detection probability parameter structures assuming Markovian emigration onl y, where transitions were state and natal river dependent (Brownie et al. 1993) This assumption was based on the h ypothesis that Gulf sturgeon summer river residence is dependent on their natal river (Wooley and Crateau 1985) supported by fidelity estimates in Chapter 2 of this study. Fitting model structures to generated data I first ran the model structures described above with generated data to ensure the multi state model could estimate unbiased parameters given limited transmitters deployed and monitoring time, un der the river specific, genetic area, and high resolution
56 genetic area spatial collapse methods. For each spatial collapse method, I conducted 1,000 iterations of data generation and respective model runs. Using this distribution of mean estimates and conf idence intervals, I calculated the number of model runs that resulted in estimation failure (confidence limits range from 0 to 1, both lower and upper confidence limits at 0 or 1, or non number confidence limits) and inaccuracy (true value used for data ge neration not in the confidence intervals estimated from generated data ). I also assessed precision of parameter estimates around the true values used to generate data Fitting model structures to real data Once I could confirm the multi state model was u nbiased, I ran all model structures described above with real data. choosing the model that could estimate survival rates and detection probability at meaningful spatiotemporal scales without failure. In addition to model comparison by AICc, models were ch osen based on the various spatial implications of the survival rate parameter structure and its ability to meet management needs (Lebreton et al. 1992) Results Simulation Test The simulation test showed that the multi state model estimates unbiased state specific survival rates and detection probability for Gulf sturgeon (Figure 3 1). With a hig h sample size of transmitters deployed, the median mean state specific survival estimate s from 1,000 iterations of data generation and model runs is equal to the true value used to generate data (Figure 3 1). Precision and accuracy of mean estimates increa ses with higher sample size of transmitters deployed and longer monitoring time (Fig ure 3 1). In accordance with this result, p recision of m arine survival rate estimates is
57 expected to be higher than river survival because transmitter sample sizes were p oo led in the marine environment (Figure 3 1 ). Riverine survival rates are estimated from individuals present in each river during the summer months, and the total number of transmitters is divided between the occupied rivers. The simulation shows that surviv al rates specific to the rivers in which fish emigrate is unlikely to be estimated with adequate precision unless a high number of transmitters are deployed in the focal river and monitored over a longer time period (Figure 3 1 ). The multi state model esti mates detection probability accurately and precisely compared to other parameters (Figure 3 1 ). The high resolution genetic area spatial collapse method was not able to converge with only two years of receiver detections. With five years of receiver detect ions, however, I would be able to estimate survival across genetic units at both riverine and marine scale (Figure 3 2). The quantified precision of model estimates will help scientists and managers understand whether Gulf sturgeon demographic population p arameter estimates from real data represent true values or are inaccurate due to low sample size or short monitoring time. In the same vein, I used the simulation to confirm that more data is necessary to accurately estimate highly parameterized models, s uch as the high resolution genetic area spatial collapse method. Although state specific survival rates can provide insight into divergent survival patterns across the Gulf over the entire monitoring period, temporal surviv al rate estimates would better indicate the timing of population impacts. Temporal survival is most precise during the middle years of monitoring (i.e. years 2, 3, and 4 of a five year monitoring program; Figure 3 3 ). Year specific survival is more precise than year and state specific survival due to the pooled sample size of deployed transmitters across all
58 rivers. Year and state specific survival would provide higher resolution into identifying the area and timing of population impacts, but it is likely implausible to estimate these parameters with reasonable accuracy and precision until five years of data are available (Figure 3 3 ). When the number of transmitters was low and monitoring time short, a higher proportion of model runs resulted in estimation failure and inaccuracy (Tabl e 3 1 ; Table 3 2 ). Under the river specific spatial collapse method, less information was available to estimate state specific survival for rivers in which fish emigrated. As a result, the probability of estimation failure or inaccuracy in survival rate es timates for emigrant ri vers was much higher (Table 3 1 ). The probability of estimation failure or inaccuracy was overall lower under the genetic area spatial collapse method than the river specific collapse method (Figure 3 1), likely a function of the poo led sample sizes by including multiple rive rs in each genetic area. The probability of estimation failure for temporal survival models was relatively high compared to other parameters with data over a two year period, but could be reasonable to compare aga inst other models aft er a five year period (Table 3 3 ). Year specific survival estimates in the marine environment are plausible over two and five years of data due to the pooled transmitter sample size from all riverine populations. Parameters Estimated from Real Data Under the river specific spatial collapse method, I chose the model with constant survival and detection probability. Constant survival for the river specific spatial collapse method is the same as natal river dependent survival, since each analysis considers individuals tagged in a single river. This model had the most AICc support for fiv e of the nine rivers (Table 3 4 ). As was indicated by high rates of estimation failure in the
59 simulation test, there was likely not enough information con tributed from rivers to which fish emigrated to get a good model fit when estimating state specific survival under the river specific spatial collapse method. Natal river specific survival rate estimates from the river specific spatial collapse method re presented an average survival across the river and marine environments for fish tagged in each respective river. The estimated survival rate for fish from the Pascagoula River was much lower than that for fi sh from other rivers (Table 3 5 ). Uncertainty sur rounding this estimate is relatively high. However, the upper 95% confidence limit is still lower than the Although the survival rate estimate in the Pearl River is on p ar with other rivers (Table 3 5 ), uncertaint y around this maximum likelihood estimate is much greater due to lower sample size of tagged individuals in that riverine population. For the genetic area spatial collapse method, I chose to compare survival rates between the models estimating state speci fic and natal area specific survival, both with state dependent detection probability. The model with natal area dependent survival received the most AICc support, while that with state dependent survival would be considered a significantly different m odel (delta AICc > 3; Table 3 6 ). The comparison of survival rates between these two models allowed me to identify differences in survival rates at the riverine and marine scale between areas in the Gulf before I have enough data to estimate sur vival on smalle r spatial scales with five years of data, either with the river specific or high resolution genetic area spatial collapse method. If the natural mortality rate for Gulf sturgeon as a species is a single unit, variable survival rates across the Gulf would i ndicate varying degrees of anthropogenic impact.
60 Under the genetic area spatial collapse method, the survival rate of Gulf sturgeon tagged in the Pearl and Pascagoula Rivers was much lower than all other rivers (Table 3 7 ). The upper 95% confidence limit on the survival rate for these fish was similar to the range of the lower 95% confidence limits for Gulf sturgeon from other rivers. Because the survival rate estimates consider survival in both the river and marine environments as one annual rate, I can a ttribute mortality to the marine and riverine habitats independently Overall, state specific survival rates indicated very low survival in the rivers of the western Gulf riverine environment compared to all others (Table 3 8 ). Survival rate estimates for the marine state and riverine environments besides the Pearl and Pascagoula Rivers were 0.85 or greater, with tighter confidence intervals around the marine survival rate estimate due to pooled transmitters in that state. Survival rate estimation in the Ch octawhatchee River resulted in failure in the state dependent survival model, likely due to issues estimating survival at the upper boundary (1.0) in the river state (Tab le 3 8 ). Under both spatial collapse methods and across all states, detection probab ility was estimated to be much higher than the conservative assumed rate utilized in data generation (Table 3 9; Table 3 10 ). Therefore, I can expect greater inference in survival rate estimates than was expected from the simulation. Under the river specif ic spatial collapse method, the estimation of detection probability at the Pascagoula River resulted in failure, likely because the model had issues estimating the parameter at its upper boundary (1.0). Detection probability was lowest on the Ochlockonee R iver, potentially because this area is believed to be a holding ground for Suwannee River Gulf sturgeon, not supporting a spawning population, so individuals may not be
61 expected to return year after year (Table 3 9). Using the genetic area spatial collapse method, the only notable difference in state dependent detection probability between models assuming state and natal river specific survival was in the western Gulf. Detection probability in the western Gulf was overall greater, with higher certainty, whe n estimated using the natal river dependent survival model (Table 3 10) Discussion Understanding landscape scale population dynamics coinciding with disturbances is necessary for conserving protected species (Do bson et al. 2001; Lowe 2002) and allows better management of species across their range A key management recommendation from the most recent Gulf sturgeon stock assessment was to develop temporal and spatial consistency in Gulf sturgeon monitoring progra ms (Pine and Martell 2009) A benefit of tagging study standardization would be the ability to compare survival rates across spatial scales. Unfortunately, my ability to make good inferences in area specific surv ival rates was limited by low sample size of deployed transmitters in the western Gulf. Using simulation based on knowledge of Gulf sturgeon ecology and population dynamics, I quantified uncertainty expected given sample size in each area and monitoring ti me. With five years of data, I can expect to make better inferences in survival rates for the eastern and central Gulf, but uncertainty in western Gulf estimates may still persist due to low sample size. Future studies and increased tagging effort are nece ssary in the western Gulf to make inferences in the regional survival rate. A unique feature of this study is the direct estimation of Gulf sturgeon natural mortality Natural mortality cannot easily be estimated by fitting a model, is difficult to estim ate externally, and can significantly bias other parameters if an uninformative prior is used (Clark 1999) Most NOAA stock assessments do not have enough data to
62 estimate natural mortality directly or cannot separate natural mortality from fishing mortality, instead relying on derivations from life history parameters (Brodziak et al. 2011) It is more conservative to assume the multi state model estimates tota l survival, translated into total mortality, to account for un quantified anthropogenic mortality. However, with area estimates from the eastern Gulf, where we expect very low frequency of i ncidental bycatch an d major anthropogenic impacts with estimates from the central or western Gulf, where anthropogenic impacts have the potential to occur at greater frequency. Overall, however, we expect mortality rates estimated from this telemetry prog ram to serve as informative priors of natural mortality for the Gulf sturgeon stock assessment. The standardized acoustic telemetry tagging program demonstrated cascading influences of higher detection probability over previous PIT tag programs that relied on physical recaptures. Spatially and technologically, acoustic telemetry is a useful tool to estimate important demographic parameters efficiently. My findings reflect those of McMichael et al. (2010) who used one tenth as many tagged juvenile salmon to estimate precise survival estimates from acoustics compared to PIT tagging methods Standardized placement of receivers at major river mouths and several smaller (Melnychuk 2009) where fishing effort may previously have been low (Fox et al. 2002) Recapture information did not require fishing effort to coincide with migration timing, which can vary by location, annual environmental stochasticity, sex, and maturity (Fox et al. 2000) The higher degree of resightings reduced confounding between survival and emigration from natal rivers (Schaub et al. 2004) Individual Gulf sturgeon
63 have many possible fates at any given time in any specific river : remaining in the riverine or marine environment off season, moving through a river branch without acoustic receivers, swimming rapidly past the acoustic receivers without detection, acoustic cancellation from boat traffic or foul weather, or mortality either in the river or marine environment. Low detection probability provides a paucity of information to make however, fewer assumptions are required and better inferences on Gulf sturgeon surviv al and movement rates can be calculated (Lebreton et al. 1992) The higher detection probabi lity associated with acoustic telemetry compared to physical recapture corroborates the relatively short monitoring period (i.e. two years) in this study Other studies have utilized the higher detection probability associated with acoustic telemetry to es timate demographic parameters over a relatively short monitoring time. Welch et al. (2009) estimated survival rates of acoustically tagged Cultus Lake sockeye salmon ( Onchorhynchus nerka ) smolts with high precision due to very high detection probability in a large scale array, despite modest sample size of tagged individuals (100 376 smolts per year for four years). My study of Gulf sturgeon tagged a similar total number of fish per year with plans to track these individuals over a similar time period, but further binne d the fish into data into smaller spatial groupings. Through simulation, Zehfuss (2000) found the number of years required to detect a trend with adequate statistic al power depended on the capture probability, initial population size, and rate of change in population size. According to her simulation, a sampling program of five years would only be justified by a capture probability of at least 0.75, a population siz e of at least 200 individuals, and a
64 relatively high rate of annual population decrease (>5%). All rivers have recent population abundance estimates greater than 200 ( Morrow et al. 1998; Rogillio et al. 2001; Berg e t al. 2007; USFWS 2007; Pine and Martell 2009; USFWS 2009) and a detection probability of 0.75 falls within the 95% confidence intervals estimated for most river mouths and genetic areas in this study. Future studies should re run the Zehfuss (2000) simulation to test for ability to estimate unbiased mortality rates for a long lived fish. Survival and detection rate estimates could be further improved with more information regarding habita t use during out migration movement. It is possible that (Melnychuk 2009) In the past, Gulf sturgeon research could only concentrate sampling effort in the areas where they knew Gulf sturgeon would be located, with the occasional trial of an alternate distributary. Acoustic re ceivers in drainages or distributaries occasionally detect Gulf sturgeon individuals. These stray detections rely on cooperation with other agencies and researchers of other species, and limited funds prevent Gulf sturgeon researchers from deploying acoust ic receivers at all possible locations. However, I can gain insights into Gulf sturgeon use of smaller distributaries from the results of varying detection probability parameter structures (Nichols and Pollock 1983) In this study, models with migration season dependent detection probability were the best fit fo r the Ochlockonee, Blackwater, and Escambia Rivers under the river specific spatial collapse method (Table 3 3). A possible reason for this model support was Gulf sturgeon out migration through smaller branches with no acoustic receivers. The Blackwater an d Escambia Rivers each only had one acoustic receiver, making it possible for Gulf sturgeon to
65 move undetected through smaller branches our out of range of the single receiver (Heupel et al. 2006) Although the Ochlockonee River had three acoustic receivers, its mouth is especially branching. There are a few possible approaches to assess the biases associated with split route out migration. In my simulation, I did not specifically account for potential movement through smaller distributary branches without acoustic receivers. Instead, this behavior would effectively decrease the detection probability. Alternatively, Melnychuk (2009) estimated bias across parameters from split route migration patterns in tagged salmon stocks using CJS a nalysis. This could be replicated within the multi state model data generation for Gulf sturgeon, and tested using receiver data. For example, the Apalachicola and Yellow Rivers each had four acoustic receivers in their associated river drainages, one at t he main mouth and three at smaller distributaries. Considering the four acoustic receivers within each river drainage as four different states, I could estimate state and migration season dependent detection probability to verify spatial and temporal diff erences in distributary use. Low CPUE in the western Gulf was reflected in the low sample size for this study, and corresponds with low estimates of abundance from previous studies (Morrow et al. 1998; Rogillio et al. 2001; Ross et al. 2001) As I demonstrated through simulation, low sample size of tagged individuals resulted in lower precision of mean survival rate estimates with simulated data and possibility of bias in survival estimates. According to Zehfuss (2000) even if the rate of population change in the western Gulf is rapid, the low population size may not be high enough for a five year monitoring program to estimate parameters with adequa te statistical power.
66 On August 13, 2011, toxic byproduct from a paper mill was spilled in the Pearl River watershed, leading to an extensive fish kill in the lower Pearl River. A total of 28 dead Gulf sturgeon were recovered by the Louisiana Department of Wildlife and Fisheries during cleanup efforts, but none had acoustic tags ( LDWF 2011) In a very small population (Rogillio et al. 2001) the fish kill would represent a very large mortality event. The fact that no recovered dead fish, as far as the LDWF knows, were killed supports the possibility that my annual survival estimate of 0.90 for fish tagged in the Pearl River is actually a natural survival estimate, as opposed to total survival. Although sample size of acoustically tagged Pearl River fish was low and uncertainty in the survival rate estimate was high, I likely have a direct es timate of natural mortality for the Pearl River sturgeon population. Issues related to low sample size of deployed transmitters in the western Gulf were also reflected in the varying estimates of detection probability in that region between state specific and natal river dependent survival models. In dealing with uncertainty, a low detection probability may lead to inaccurate survival rates and less precision, while a higher capture probability would result in more accurate survival rate estimates with mor e precision (Nichols and Pollock 1983) The difference in detection probability and survival rate estimates between models in the western Gulf is an example of the parameter tradeoff and uncertainty associated with a small tagged population. Higher detection probability in the natal river dependent survival mode l potentially resulted in a higher survival rate estimate and lower uncertainty, but was confounded with low sample size and the different spatial scales represented in each
67 survival rate parameter structure. More data is needed to relieve biases in all pa rameter estimates. My ability to identify population impacts and estimate parameters at meaningful spatial scales is hindered if my assumptions of zero tag loss and tag mortality are violated. Field biologists tracked two acoustic tags in the Choctawhatc hee River during November and December 2012 after fish should have migrated out into the bay. One of these tags was also tracked in nearly the same area during winter 2011 (D. Fox and N. Willett, Delaware State University, personal communication). Matching these observations with the telemetry data, neither of these transmitters was detected at the river mouth after they were surgically implanted. However, it is not possible to discern whether these occurrences represent tag loss or mortality with this info rmation only. If mortality occurred after tagging, this would likely be reflected in the survival rate estimation due to the chain of non detections in the capture history. The results of this study, combined with those of Chapter 2, have significant man agement implications for Gulf sturgeon under ESA legislation. First and foremost, these initial data indicate the survival rate of Gulf sturgeon in the Pearl and Pascagoula Rivers is lower than all other habitats (rivers and marine). This is reflected in t he highly variable and relatively lower numbers in abundance reported (USFWS and NMFS 2009) In fact, events from the black liquor event tha t occurred on the Pearl River were undetectable by this model given the small number of tagged individuals. The purpose of the tagging program was to gain better inferences in Gulf sturgeon survival and movement rates. The program was successful on these c ounts, in central and eastern Gulf of Mexico populations. Although a limitation of the tagging program was the low
68 sample size of tagged individuals in the western Gulf populations, this serves as a warning sign that these populations may be in more danger Managers should act quickly and look at river specific threats in response to low CPUE and abundance estimates in the Pearl and Pascagoula Rivers. The data indicate trouble, but pre exploitation and current carrying capacity estimates would provide addit ional insight into the population sizes expected in western Gulf riverine habitats. U pdating recovery criteria and utilizing the wide knowledge base on eastern populations will begin the process towards species recovery and reallocation of funds towards ar eas in need.
69 Table 3 1. Proportion of estimation failure and confidence intervals not including true parameter values for state specific survival parameters and constant detection probability using the river specific spatial collapse method Estima tion Failure True Value Not In CI # Transmitters Deployed Low Medium High Low Medium High 2 YEARS Focal River Survival 0.14 0.12 0.12 0.52 0.47 0.43 Marine Survival 0.09 0.08 0.08 0.40 0.37 0.23 Emigrated River Survival 0.64 0.56 0.38 0. 93 0.86 0.69 Detection Probability 0.01 0.00 0.00 0.06 0.08 0.09 5 YEARS Focal River Survival 0.10 0.10 0.06 0.33 0.28 0.16 Marine Survival 0.09 0.08 0.04 0.16 0.12 0.03 Emigrated River Survival 0.30 0.22 0.18 0.68 0.58 0.38 Detec tion Probability 0.00 0.00 0.00 0.05 0.04 0.05 Note: O ut of 1,000 iterations of generated data and model runs for a range of sample sizes of deployed transmitters mo nitored over two and five years. Low, medium, and high refer to the relative number of t agged individuals in each river. river into which the tagged fish move that is not their natal river.
70 Table 3 2. Proportion of estimation failu re and confidence intervals not including true parameter values for state specific survival parameters and constant detection probability using the genetic area spatial collapse method. Estimation Failure True Value Not In CI # Transmitters Deployed Low Medium High Highest Low Medium High Highest 2 YEARS Genetic Area Riverine Survival 0.18 0.14 0.14 0.14 0.37 0.34 0.34 0.30 Marine Survival 0.01 0.11 Detection Probability 0.00 0.05 5 YEARS Genetic Area Riverine Survival 0.0 4 0.03 0.03 0.01 0.17 0.14 0.12 0.12 Marine Survival 0.00 0.06 Detection Probability 0.00 0.06 Note: O ut of 1,000 iterations of generated data and model runs for a range of sample sizes of deployed transmitters mo nitored over two and five years. Low, me dium, high, and highest refer to the relative number of tagged individuals in each river.
71 Table 3 3 Proportion of estimation failure and confidence intervals not including true parameter values for year specific and year and state specific survival r ates 2 years 5 years Estimation Failure True Value Not in CI Estimation Failure True Value Not in CI Yr 1 0.03 0.02 0.00 0.13 Yr 2 0.00 0.11 0.00 0.18 Yr 3 0.02 0.38 0.00 0.21 Yr4 0.00 0.10 Yr5 0.00 0.06 Low Yr1 0.97 0.53 0.26 0.16 Low Yr2 0.38 0.37 0.12 0.16 Low Yr3 0.96 0.59 0.17 0.16 Low Yr4 0.18 0.17 Low Yr5 0.14 0.13 Medium Yr1 0.97 0.46 0.11 0.14 Medium Yr2 0.33 0.34 0.06 0.18 Medium Yr3 0.96 0.57 0.08 0.17 Medium Yr4 0.06 0.29 Medium Yr5 0. 03 0.19 High Yr1 0.98 0.45 0.12 0.10 High Yr2 0.31 0.31 0.06 0.12 High Yr3 0.97 0.54 0.07 0.14 High Yr4 0.07 0.16 High Yr5 0.05 0.16 Highest Yr1 0.99 0.41 0.10 0.11 Highest Yr2 0.27 0.23 0.06 0.14 Highest Yr3 0.95 0.53 0.05 0 .11 Highest Yr4 0.07 0.12 Highest Yr5 0.05 0.13 Pooled Marine Yr1 0.10 0.08 0.04 0.12 Pooled Marine Yr2 0.07 0.12 0.06 0.13 Pooled Marine Yr3 0.02 0.23 0.05 0.11 Pooled Marine Yr4 0.03 0.19 Pooled Marine Yr5 0.17 0.28 Note: O ut of 1,000 iterations of generated data and model runs using the genetic area spatial collapse method for a range of sample sizes of deployed transmitters monitored over two and five years. Low, medium, high, and highest refer to the relative numb er of tagged individuals in the group of rivers within a genetically distinct unit.
72 Table 3 4 AICc table comparing models of Gulf s turgeon population dynamics for real data under the river specific spatial collapse method Note: on dependen epresents constancy across states, years, and groups. Model Lowest delta AICc S(. )p(.)Psi(s) Suwannee, Apalachicola, Yellow, Pascagoula, Pearl S(.)p(m)Psi(s) Ochlockonee, Blackwater S(y)p(.)Psi(s) Choctawhatchee S(y)p(m)Psi(s) Escambia
73 Table 3 5. Gulf sturgeon i nstan taneous total survival estimates using the river specific spatial collapse method for river specific rates, with upper (UCL) a nd lower (LCL) 95% confidence limits. River Estimate LCL UCL Pearl 0.90 0.61 0.98 Pascagoula 0.51 0.34 0.66 Escambia 0.95 0.82 0.98 Blackwater 0.87 0.73 0.95 Yellow 1.00 0.00 1.00 Choctawhatchee 0.98 0.88 0. 99 Apalachicola 0.89 0.69 0.97 Ochlockonee 0.72 0.53 0.86 Suwannee 0.97 0.84 0.99
74 Table 3 6 AICc table comparing models of Gulf s turgeon population dynamics for real data under the genetic area spatial collapse method Note: dependent, dependent, e presents constancy across states, years, and groups. An asterisk denotes an interaction, such dependent. Model delta AIC S(g)p(s)Psi(s*g) 0.0 S(g)p(m)Psi(s*g) 1.2 S(s)p(s)Psi(s*g) 8.4 S(g)p(.)Psi(s*g) 9.4 S(s)p(m)Psi(s*g) 10.8 S(g)p(s*m)Psi(s*g) 10.9 S(.)p(s)Psi(s*g) 11.8 S(y)p(s)Psi(s*g) 13.2 S(. )p(m)Psi(s*g) 13.3 S(y)p(m)Psi(s*g) 14.2 S(s*y)p(s)Psi(s*g) 14.7 S(s)p(.)Psi(s*g) 16.7 S(s*y)p(m)Psi(s*g) 16.8 S(y)p(.)Psi(s*g) 18.5 S(s)p(s*m)Psi(s*g) 19.4 S(s*y)p(.)Psi(s*g) 19.7 S(.)p(.)Psi(s*g) 21 .7 S(.)p(s*m)Psi(s*g) 22.6 S(y)p(s*m)Psi(s*g) 24.1 S(s*y)p(s*m)Psi(s*g) 25.9
75 Table 3 7. Gulf sturgeon i nstan taneous total survival estimates using the genetic area specific spatial collap se method for natal area specific rates, with upper (UCL) and lower (LCL) 95% confidence limits. Genetic Area Estimate LCL UCL West 0.70 0.55 0.82 Escambia Bay 0.92 0.85 0.96 Choctawhatchee 0.98 0.86 0.99 East 0.92 0.81 0. 97
76 Table 3 8. Gulf sturgeon i nstan taneous total survival estimates using the genetic area specific spatial collapse method for state specific rates, with upper (UCL) and lo wer (LCL) 95% confidence limits. Genetic Area State Estimate LCL UCL West 0.61 0.34 0.82 Escambia Bay 0.96 0.64 0.99 Choctawhatchee 1.00 0.99 1.00 East 0.85 0.68 0.94 Marine 0.93 0.86 0.97
77 Table 3 9 River specific d etection probability estimates with upper (UCL) a nd lower (LCL) 95% confidence limits. River Estimate LCL UCL Pearl 0.73 0.46 0.89 Pascagoula 1.00 1.00 1.00 Escambia 0.68 0.56 0.79 Blackwater 0.66 0.51 0.78 Yellow 0.55 0.46 0.64 Choctawhatchee 0.68 0.60 0.75 Apa lachicola 0.79 0.58 0.91 Ochlockonee 0.55 0.22 0.84 Suwannee 0.94 0.83 0.98
78 Table 3 10. D etection probability estimates using the genetic area spatial collapse method from models assuming state specific and natal river specific surviva l, with upper (UCL) and lower (LCL) 95% confidence limits. State specific survival model Natal river dependent survival model Genetic Area State Estimate LCL UCL Estimate LCL UCL West 0.65 0.48 0.79 0.88 0.55 0.97 Escambia Bay 0.71 0.62 0.79 0 .72 0.62 0.8 Choctawhatchee 0.71 0.61 0.79 0.68 0.58 0.76 East 0.85 0.74 0.92 0.85 0.71 0.93 Marine 0.62 0.54 0.69 0.62 0.54 0.69
79 Figure 3 1. Box and whisker plot s of the distribution of mean river specific survival and detection probability estimates representative for Gulf sturgeon from 1,000 iterations of generated data A) Survival rate in the river in which the Gulf sturgeon was hypothetically tagged. B) Survival rate in a river to which a Gulf sturgeon hypothetically emigrate s. C) Survival rate in the marine environment. D) Detection probability. Precision of estimates are compared between two and five years of generated data and a range of sample sizes of tagged Gulf sturgeon. Shaded areas rtile range, whiskers extend to the most extreme data point no longer than range multiplied by the interquartile range, and indivi dual points represent outliers. Colored lines indicate the true values of Gulf sturgeon survival and detection probability us ed in data generation. A B C D
80 Figure 3 2 Box and whisker plot of the d istribution of mean genetic area specific riverine and marine survival rates using the high resolution spatial collapse method applied to 1,0 00 iterations of generated data. Generated cap ture histories represent five year of data with a range of sample sizes of tagged Gulf sturgeon Shaded areas whiskers extend to the most extreme data point no longer than range multiplied by the interquart ile range, and individual points represent outliers. Blue and red lines indicate the true values of Gulf sturgeon marine and riverine survival, respectively, used in data generation.
81 A Figure 3 3. Distribution of me an total year specific survival an d year and state specific survival estimates from 1,000 iterations of generated data A) Generated data represents detection histories over two years. B) Generated data represents detection histories over five years. Generated data includes a range of sam ple sizes of tagged Gulf sturgeon (i.e. low, medium, and high). There are three year specific survival estimates when data was generated over the first two years of data because the spring season (March May) is considered part of the next year. Shaded ar eas range, whiskers extend to the most extreme data point no longer than range multiplied by the interquartile range, and individual points represent outliers. Purple, red, and blue lines indicate the true values o f Gulf sturgeon total, riverine, and marine survival, respectively, used in data generation.
82 B Figure 3 3 Continued.
83 CHAPTER 4 STOCK ASSESSMENT OF GULF STURGEON USING AGE STRUCTURED MARK RECAPTURE ANALYSIS Evaluation of stock status is vital for ra re or endangered species whose abundance is often not precisely known, but the uncertainty warrants protective status mandated by state or federal regulations (i.e. Magnuson Stevens Fisheries C onservation and Management Act) (Pollock et al. 2002) For these species, managers often infer stock status from trends in catch per unit effort (CPUE) indices or capture recapture methodologies. Estimates of trends in abundance and mortality are used to determine whether management actions are necessary to reduce likelihood of population extinction. Estimating abundance of Gulf sturgeon populations has been a primary motivation for many previous studies on this species ( Wooley and Crateau 1985; Chapman et al. 1997; Sulak and Clugston 1999; Zehfuss et al. 1999 ; Pine et al. 2001 ) Tagging data are not continuous for any system, but the Suwanne e and Apalachicola Rivers have received the largest amou nt of sampling effort (Table 4 1) Portions of these data sets have been used previously to determine population viability (Pine et al. 2001) estimate abundance (Sulak and Clugston 1999) and provid e guidance on Gulf sturgeon sampling programs (Zehfuss et al. 1999) Published estimates for Gulf sturgeon abundance generally range from severa l hundred individuals in the Apalachicola River (closed models) (Zehfuss et al. 1999) to 5,000 7,000 individuals in the Suwannee River (open a nd closed models) (Chapman et al. 1997; Sulak and Clugston 1999) However, biases are possible in these methods due to violations of model assumptions (Pollock et al. 1990) Both model types assume homogeneous capture probabilities, either for all animals in the system at all time periods ( closed models) or during any given time (open models).
84 In the case of Gulf sturgeon and many fisheries applications, heterogeneous capture probabilities are related to individual age. To deal with this issue for the endangered humpback chub ( Gila cypha ), Coggins et al. ( 2006 a; 2006b) developed an age structured mark recapture (ASMR) analysis that combines attributes from the classic Jolly Seber open population mark recapture models and virtual population analysis (VPA) methods. A key attribute of this approach is that the ASMR analysis estimates the number of individuals expected to be caught based on the number at risk (predicted by the age structured VPA) and capture probability of previously marked fish (from th e Jolly Seber model). The ASMR analysis attempts to reduce uncertainty in demographic parameters by treating unmarked fish at age and new recruits entering the unmarked population as known parameters (Coggins et al. 2006b) This approach differs from traditional capture recapture approaches that primarily use information on recaptures of previous marked indiv iduals (Williams et al. 2001) Especi ally for rare or protected species where lethal but more precise aging methods are not appropriate, there may be high variability in the age of a fish of a particular length. This is especially an issue for long lived species such as Gulf sturgeon, as an i ndividual could be a wider range of ages at longer lengths. Uncertainty in initial age assignment can propagate through to the estimation of year specific values of parameters of interest (e.g. survival abundance, and recruitment). Coggins (2007) incorporated uncertainty from initial age assignment into the humpback chub assessment and found that estimated adult abundance is still precise, yet estimates of recruitment were relatively imprecise. The rigorous methodology of reiterating the age assignment based on length is an important step to understand uncertainty around
85 recruitment time series estimates especially, for the potential identification of significant population impact. The overall objective of this chapter is to estimate current and historical abundance of Gulf sturgeon for Florida waters off the Gulf of Mexico. I estimate population size, age 1 recruitment, and natural mortality by using ASMR analysis (Coggins et al. 2006b) We use data sets from the Suwannee and Apalachicola Rivers, which date back the longer than other systems (late 1970s to present). The Apalachicola and Suwannee Rivers represent systems where population abundance was thought to be low (Apalachicola River) and high (Suwannee River). This contrast in abundance would allow for comparison of population model performance with data from populations of different sizes. By estimating the current, and reconstructing the historical population size of this species, I attempt to place the current abundance of this species in Florida in a comparative context to assess relative population status from a conservation perspective. I provided hypotheses on potential factors driving observed patterns in these Gulf sturgeon po pulation vital rates, and close with recommendations for changes in field programs to improve future assessment activities by reducing uncertainty. Methods Historic Mark Recapture Database The 2009 Gulf sturgeon stock assessment called for a standardized centralized database and sampling coordination office for archiving historical data collection efforts. In response, The NOAA Southeast Fisheries Science Center in Panama City, Florida developed an online database to house mark recapture information for each individual tagged in any river system from 1977 to present. A database user can filter information
86 based on river, year, PIT tag or left or right T bar tag. Information for each capture occasion includes tag number, animal ID, occurrence, total length weight, gill net type (anchored or drift), and other factors. Over time, some Gulf sturgeon accrued several tags, either due to double tagging to protect against tag loss, unidentified PIT tags, or standardization of PIT tag frequencies in recent years. Data managers matched T bar, PIT, and acoustic tags to assign each individual an animal ID number to more easily track individuals over time. Occurrence specified the number of times each individual was caught. Therefore, an occurrence of 1 is an initial c apture, and an occurrence greater than 1 is a recapture. Gaps in the data exist for some years due to limited funds for field sampling efforts. A few other data errors exist, including missing total length values. I exported data for the entire available p eriod for the Suwannee and Apalachicola Rivers (Figure 2 1 ). I used these data to assess trends in abundance, recruitment, and mortality using ASMR methods. The Suwannee and Apalachicola Rivers have been sampled for the longest time period with the largest effort and cumulative tagged populations Sampling in the Apalachicola is slightly more episodic than the Suwannee River. Initial Age Assignment Limited aging information is available for Gulf sturgeon due to their protected status. Within the NOAA data base, ages for 487 individuals are available based on estimates from pectoral fin rays from the Suwannee, Apalachicola, and Yellow Rivers ( Huff 1975 ; Berg 2004 ) Due to limited data, I ass umed growth rates were similar between rivers and I fit a von Bertalanffy growth curve to the available age length information by the equation:
87 ( 4 1) where is mean predicted length at age, is asymptotic length for the population, is the Brody growth coefficient, is age, and is the theoretical age at length 0. A likelihood function based on the normal probability distribution minimized the deviation between predicted and observed length for each aged individual (x): ( 4 2) by varying a coefficient of variation ( CV ), , and where was the mean length at age and was multiplied by CV Markov Chain Monte Carlo (MCMC) methods, using the Random Walk Metropolis Hastings in AD Model Builder (Fournier et al. 2012) constructed a Markov chain to sample the probability distribution and produce a posterior distribution for the von Bertalanffy growth parameters. I used ASMR analysis to calculat e river specific estimates of population size, natural mortality, capture probability, and recruitment. The data structure for this method requires two matrices: initial captures at age over time and recaptures at age over time. The dataset was split into initial captures (occurrence = 1, first observation of the fish in the dataset) and recaptures (occurrence > 1). For each individual initially captured, I assi gned an age based on its length. Ages were assigned based on the proportion of population at each age : ( 4 3) where is survival at age based on the Lorenzen age specific survival equation (Lorenzen 2000) M is natural mortality, and is survivorship at age. I then used the
88 normal probability density function (E quation 4 2 ) to calculate the cumulative probability of an individual being an age a given length L: (4 4 ) I then drew a random number from a uniform distribution between 0 and 1 and chose the first possible age where the random number was less than or equal to the cumulative probability of being that age given length. If the random number is high, the age assignment will be on the older side of possible ages given length. With a lower random number, the fish could be a wider range of ages but will be assigned an age on the younger range. Initial age assignment is important because if uncertainty exists in the age assignment it can propagate through to the estimation of year specific values of parameters of interest (e.g. survival abundance, and recruitment). For many species, particularly long lived species such as sturgeon, age uncertainty can contribute great uncertainty into the age distribution (in turn, stock assessment parameters) due to the wide range of ages an individual could be at longer lengths (Figure 4 2 ). In some cases, total length was not available upon initial capture. Individuals without length at capture that were not later recaptured were not included in this analysisf For individuals with no length information at initial capture but with length at recapture, I assigned an age based on its length at second capture and then back calculated the age at fir st capture as the number of years between the initial capture and recapture. Age Structured Mark Recapture Model The ASMR model worked at an annual time step, such that the marks and recaptures required were also annual. I matched the year each fish was initially caught
89 with its assigned age to create a matrix denoting initial captures at age over time. I created a separate matrix for recaptures at age over time by matching animal ID numbers for recaptured individuals with their initial capture informatio n. I then assigned each recaptured individual an age based on the number of years that passed between initial capture and recapture. Gulf sturgeon captured multiple times in one year were not considered separate recaptures. Coggins et al. (2006b) explored three assumptions of ASMR reflecting different ASMR 1 and 2 models estimate predicted number of marks and recaptures as a function of time dependent capture probability Alternatively, ASMR 3 estimates predicted marks and recaptures as a function of observed number marks and recaptures and the estimate d marked and unmarked population sizes, not incorporating capture probability. The ASMR 1 and 2 models differ by the direct estimation of either terminal capture probability (ASMR 1) or terminal age specific abundance of the unmarked population (ASMR 2), w hich in turn affects the back propagation to estimate the unmarked population size over time (Coggins et al. 2006b) I chose to assess Gulf sturgeon riverine populations using ASMR 2, assuming capture probability is a factor in the predicted number of marks and recaptures per year based on the variable sampling over time. Additionally, I would expect higher certainty in my estimate of the total unmarked population size by utilizing age specific survival in the direct estimation of the unmarked population size at age in the terminal year, as opposed to an indirect estimation as a function of capture probabilit y and vulnerability at age. Analysis of ASMR 2 consists of a series of major functions, including (1) estimating the population size of unmarked and marked fish, (2)
90 estimate capture probabilities and predicted number of marks and recaptures, (3) calculate the negative log likelihood, (4) estimate recruitment, and (5) run MCMC. The first major ASMR function is the estimation of unmarked and marked fish abundances. The estimation of these parameters depends on the estimation of age specific survival (Lorenzen 2000) and vulnerability at age based on a cumulative logistic function. The natural mortality rate of adults (a parameter in the age specifi c survival function), the 50 th percentile of individuals vulnerable to the gear (logistic curve inflexion point), and the standard deviation of the logistic distribution of the vulnerability schedule are all parameters estimated in MCMC analysis (Fournier et al. 2012) The ASMR model estimated the expected number of unmarked ( ) and marked ( ) fish, by age ( ) and year ( ): ( 4 5) ( 4 6) where is the number of age a fish marked in year t Equations 4 5 and 4 6 assume all unmarked fish captured received a tag. Using the VPA backward propagation fram ework, I assumed the unmarked population at the oldest age A would be 0 for all years t given the oldest possible age A is beyond the oldest age the fish can attain. I used a terminal age of 50, similarly to Flowers et al. (2009) based on estimates of longevity (Hewitt and Hoenig 2005) and several captures of Gulf sturgeon at large for more than ten years between recaptures and initial age assignment up to 25 years old. The only unknown param eter for estimation is the number of unmarked fish at age in the terminal year,
91 The next major ASMR function is the estimation of time dependent capture probability : ( 4 7) where is the number of age a fish recaptured in year t and is the vulnerability to the gear at age over time. Time dependent capture probability is then used to estimate the predicted numbe r of marks ( ) and predicted recaptures ( ) at age over time: ( 4 8) ( 4 9) Next, I assumed a negative binomial distribution for the data, l eading to the log likelihood function: ( 4 10) where is the parameter vector to be estimated, inc luding vulnerability at age over time ( ), adult natural mortality rate ( ), and terminal age specific unmarked population size ( ) The ASMR 2 model maximizes Equation 4 10 by varying theta The ASMR model estimates time dependent recruitment of age 1 Gulf sturgeon in the VPA back calculation of the unmarked population and ten years preceding data collection using the equation: ( 4 11) which depends on age sp ecific survival estimates and estimates of the unmarked population in the first year of data collection (Coggins et al. 2006b).
92 To better quantify uncertainty in parameter estimates, I utilized the random walk Metropolis Hastings algorithm in AD Model Bu ilder to integrate the posterior distributions for the natural mortality rate of adults, the 50 th percentile of individuals vulnerable to the gear (logistic curve inflexion point), the standard deviation of the logistic distribution of the vulnerability sc hedule, the abundance of the population of age 2 and older Gulf sturgeon each year, and the number of age 1 recruits each year (Fournier et al. 2012) For the natural mortality rate and parameters in the vulnerability curve, I set priors based on previous studies contributing to knowledge on Gulf sturgeon life history (e.g. M=0.08 ). I used the maximum likelihood estimates (MLEs) of the abundance of Gulf sturgeon age 1 recruits and age 2 and older as priors for MCMC evaluation. To facilitate analysis and plotting, I sampled every 50 th trial from the MCMC posterior chain of length 10,000 discarding the first half of the chain as burn in This number of samples was sufficient according to convergence criteria in Gelman et al. (2000) To incorporate uncertainty in age assignment at init ial capture, I iterated ASMR methods and MCMC analysis 2,000 times for each river, each iterate with a separate age assignment based on length according to the normal likelihood distribution (Equation 4 1). To incorporate uncertainty in initial age assignm ent, I combined the MCMC chains from each iterate of age assignment. I constructed 95% confidence interval s from the total MCMC chains concatenated to better characterize uncertainty in these vital parameter estimates (Gelman et al. 2000) Results Data Export Mark recapture data for Gulf sturgeon was available from the S uwannee Rive r from 1982 to 2012 and the Apalachicola River from 1977 to 2012 (Table 4 1) Detailed
93 information on sampling approaches is available from previous studies in these rivers (Fox et al. 2000; Sulak and Clugston 1999; Wooley and Crateau 1985; Zehfuss et al. 1999) Low marking and recapture effort existed in the database for the Suwannee River from 2007 to 2012 (Table 4 1), resulting in lack of model convergence when those years were included. Therefore, I excluded Suw annee River data after 2007 for this study. Individuals that were recaptured that had not been originally tagged in the river of interest were excluded from this analysis (Table 4 2). Sampling on the Suwannee River was fairly constant over time, while mark ing and recapture effort was relatively more episodic on the Apalachicola River (Figure 4 1). We determined visually that the model converged given the described sampling from the MCMC posterior distribution. Parameter Estimation Natural mortality Evalua ted using MCMC methods, natural mortality rate estimates for both the Apalachicola and Suwannee Rivers were lower than previous estimates for Gulf sturgeon throughout their range ( Zehfuss et al. 1999 ; Flowers 2008 ) The median natural mortality rate estimates from the MCMC posterior distribution for the Apalachicola and Suwannee Rivers were 0.032 and 0.031 respectively (Table 4 3). Their constructed 95% c onfidence intervals were narrow, suggesting good parameter fi t for both rivers (Table 4 3). Abundance of age 2+ and 4+ individuals Median estimates and 95% confidence intervals of Gulf sturgeon population abundance of fish age 2 and older were constructed from MCMC analysis, incorporating all iterations of random a ge assignment. Estimates of population abundance show a relatively stable population trajectory in the Suwannee River and upward population
94 trajectory in the Apalachicola River (Figure 4 3 ). I estimated the median terminal abundance of Gulf sturgeon age 2 and older in the Suwannee River from the posterior distribution to be about 12,100 individuals in 2007, with a lower 95% confidence limit of 10,450 Gulf sturgeon and upper 95% confidence limit of 13,900 Gulf sturgeon (Figure 4 3) My estimated median termi nal abundance of Gulf sturgeon age 2 and older in the Apalachicola River in 2012 was about 2,450 individuals, with a lower 95% confidence limit of 1,960 Gulf sturgeon and upper 95% confidence limit of 3,170 Gulf sturgeon (Figure 4 3). Alternatively, in th e maximum likelihood estimation of population abundance to produce the prior for MCMC analysis, the population trajectories from 2,000 iterations of age assignment were bimodal (Figure 4 4 ) Out of 2,000 iterations of random age assignment, 77 maximum like lihood estimates of the Apalachicola River Gulf sturgeon population abundance were declining over time. However, when used as a prior in the MCMC analysis, the posterior distributions estimated increasing population trajectories. This means that the MCMC a nalysis estimated robust measures of population abundance over time, contrasted with the maximum likelihood analysis occasionally getting stuck at an alternate area of the probability distribution. Even with this assumed uninformative prior (declining popu lation trajectory), MCMC analysis found the solution that agrees with the majority of the MLEs (Figure 4 4 ). Age 1 Recruitment Median estimates and 95% confidence intervals of Gulf sturgeon age 1 recruitment were constructed from MCMC analysis, incorporat ing all iterations of random age assignment (Figure 4 5). Recruitment estimates in the Suwannee River slowly increases from around 390 age 1 Gulf sturgeon in 1972 to around 1,290 recruits
95 in 1990. Recruitment in the Suwannee River reaches a peak in 1991 at about 3,170 age 1 Gulf sturgeon, but promptly drops down to previous levels for the remainder of the time period. This peak occurs the year prior to the large spike in Suwannee River Gulf sturgeon population abundance of age 2 and older (Figure 4 3). Unce rtainty, expressed as 95% confidence intervals, remains relatively stable throughout the time period (Figure 4 5). A year class must be fully recruited to the capture gear for three or four years before it is possible to assess the relative strength of the year class (Coggins et al. 2006a) This is due to the nature of time dependent estimates using capture recapture methods, selectivity patterns of the gear, and potentially issues with the VPA estimating unb iased parameters for recent years when combined with low captu re probability. Therefore, we only consider recruitment estimates through 2003 in the Suwa nnee River for this assessment, where the median estimate from the posterior distribution was approximat ely 320 Gulf sturgeon age 1 recruits. In the Apalachicola River, annual age 1 recruitment of Gulf sturgeon is estimated to be low, slowly increasing from a median estimate of 27 age 1 Gulf sturgeon in 1967 to about 150 recruits in 1996. After 1996, age 1 r ecruitment of Gulf sturgeon in the Apalachicola River becomes more variable, with a peak of 220 age 1 Gulf sturgeon in 2001 and a relatively steady, steep increase from 2003 to 2008 when the median age 1 recruitment estimate was 370 Gulf sturgeon Also du e to spurious recruitment estimates for the terminal four years of data collection, I only consider recruitment estimates through 2008 in the Apalachicola River for this assessment. Uncertainty in the recruitment estimates from MCMC, incorporating uncertai nty in initial age assignment,
96 remains relatively stable throughout the time period but increases slightly in more recent years associated with an increasing trend in estimated recruitment. Capture probability The median estimates and 95% confidence inter vals for capture probability each year were constructed from the distribution of mean capture probability MLEs from 2,000 iterations of random age assignment. Median c apture probability of Gulf sturgeon in the Suwannee River from the distribution of MLEs i ncorporating uncertainty in initial age assignment varied between 0.00 and 0.16 over the time series reaching a peak in 1992 and dropping very low after 1995 (Figure 4 6 ). Median c apture probability from the distribution of MLEs with different iterations of age assignment in the Apalachicola River remained in the same range as those in the Suwannee River, varying b etween 0.00 and 0. 14 (Figure 4 6 ). However, f or the iterations of age assignment estimating a downward population abundance trajectory after the mid 1990s, capture probability MLEs ranged between 0.00 and 0.39 outside of the 95% confidence intervals (Figure 4 7 ). Exploring bifurcation in abundance estimates Years when the capture probability MLEs were estimated with higher uncertainty associated with initial age assignment occur in years when the MLEs for population abundance in the Apalachicola River become divergent (Figure 4 7). This became a clue in understanding why a non negligible number of iterations of age assignment resulted in declinin g maximum likelihood estimates of abundance from 1997 2012. The same iterations of age assignment that resulted in declining population trajectories also estimated MLEs of capture probability and natural mortality as higher and more uncertain (Figure 4 7). The average natural mortality MLEs for iterations with increasing popu lation trajectories was 0.0 3 while those associated with decreasing population
97 trajectories was 0.05. It is likely that the maximum likelihood framework was estimating one of these rel ated parameters in an alternate area of the probability distribution. As an alternative formulation of the likelihood term, I re fit the ASMR model to the Apalachicola River data, but this time assumed a Poisson probability distribution. Using a Poisson instead of a negative binomial likelihood did not result in bifurcation in the abundance tr ajectories over time (i.e. abundance estimates over time for each iterate of age assignment had approximately the same, positive slope) To better understand the di fferences between the probability distributions that led to divergent results, I compared the probability distributions of predicted recaptures of Gulf sturgeon during 2004, when the abundance trajectories assuming a negative binomial distribu tion were bif urcated (Figure 4 8 ). A key difference in these two likelihood formulations is the Poisson distribution assumes the mean and variance are equal, while the negative binomial incorporates a dispersion parameter ( k ) to describe the relationship between varian ce and the mean. The Poisson distribution is the limit of the negative binomial distribution; as k approaches zero, the two distributions are equal. The median of the MCMC posterior distribution for k in the Apalachicola River was 12.2 while that for the Suwannee River (where the negative binomial distribution appeared more robust to iterations in random age assignment) was estimated as 4.4 This would explain why it was less likely to see bifurcation in maximum likelihood estimates in the Suwannee River a ssessment than for the Apalachicola River Gulf sturgeon population. Discussion Using long term mark recapture datasets for Gulf sturgeon from two rivers in Florida, I found that Gulf sturgeon abundance was generally increasing or constant over the past 2 5 years I characterized how a key assumption of my models, correct
98 assignment of age at initial capture, could influence results. I found that estimates of my key parameters, such as abundance and mortality, are likely influenced more by changes in sampli ng programs and associated changes in capture probability, than in mis assignment of age at first capture. These results are of significant interest to resource managers charged with assessing Gulf sturgeon stock status in the Gulf of Mexico. My results su ggest that while Gulf sturgeon stock status appears to be improving, at least in the river systems evaluated, variation in capture probability likely driven by changes in sampling programs may reduce the inference possible from the extensive passive fish s ampling programs. The long term nature of the historic mark recapture database of Gulf sturgeon initial marks and recaptures proved to be beneficial in resolving uncertainty in natural mortality estimates. Several Gulf sturgeon were recaptured ten or more years after previous capture in the Apalachicola and Suwannee Rivers. These recaptures were very informative for the natural mortality rate estimation, and resulted in rates expected for a long lived fish and similar to those for other sturgeon population s ( Kahnle et al. 1998 ; Irvine et al. 2007 ) In studies over shorter periods, survival rate estimates would have lower certainty due to the unknown fate of sturgeon after several seasons of non detection, especially if capture probabilities are low as is commonly observed in passive tagging studies (Pine et al. 2003) The telemetry study presented in Chapters 2 and 3, for example, estimated total mortalit y for the Suwannee and Apalachicola Rivers as 0.030 and 0.117 respectively (Chapter 3). However, even though the capture probability estimates using the virtual reca ptures of telemetered animals was much higher (generally 0.60 0.85) than those observed for passive tags (generally 0.05 0.15), the
99 short term duration of the telemetry study (only three years of data available of a five year study) presents a limitation i n estimating survival rates for sturgeon with adequate certainty. I found that uncertainty around survival and abundance estimates using telemetry was higher than that from the posterior distributions from MCMC in the ASMR model, even when I incorporated a relatively high level of uncertainty in age assignments. For all parameters estimated, the uncertainty in the parameter estimates was highest in the Apalachicola River, likely because of the lower cumulative tagged population and the tagging data were col lected with short periods of relatively intensive sampling followed by relatively long periods of limited sampling effort. Coggins and Walters (2009) presented a series of retrospective analyses for humpback chub using ASMR where they assess how est imates of survival change as additional years of sampling take place. These authors found that for a long lived species, adult mortality rate estimates declined for a given year and age class of tagged Gulf sturgeon as those fish were recaptured in future years. My mortality estimates for Gulf sturgeon from pas sive tagging, about 0.03 are lower than previous studies, which ranged from 0.08 to 0.16 in the Suwannee and Apalachicola Rivers ( Sulak and Clugston 1999; Ze hfuss et al. 1999 ; Pine et al. 2001 ; Flowers 2008 ) These previous estimates were either based on life history parameters, or used portions of the same mark recapture dataset from this study (i.e. a shorter time period than this study). The results of ASM R incorporating uncertainty in initial age assignment provide relatively certain mortality estimates for Gulf sturgeon populations in the Apalachicola and Suwannee Rivers, and serve as a prime example for long term sampling programs to extend into other Gu lf sturgeon riverine populations.
100 Abundance estimate s from the ASMR model are similar to those from previous studies but with higher certainty due to model assumptions Previous studies estimated abundance from Lincol n Petersen like methods and Cormack Jolly Seb er (CJS) mark recapture models. Researchers from USGS used the CJS mark recapture model with historic mark recapture data from 1986 to 2007 for adults and sub adult Gulf sturgeon assuming constant survival and time dependent capture probability ( M. Randall USGS, unpublished data). These data are likely the same as those centralized database in coinciding years but it is possible the USGS CJS model utilized more Suwannee River data than was available in the NOAA centralized da tabase The USGS analyses found patterns in time dependent capture probability about the same as those estimated by ASMR in this study, but were greater in some years (Figure 4 9 ). In 2006, when the CJS model estimated a high abundance of 14,500 Gulf sturg eon corresponding with a capture probability of 0.03, the ASMR model estimated the same capture probability, but an estimated abundance of 12,500 individuals. The estimated mortality rate from the CJS model was 0.12, much greater than my estimated natural mortality rate of 0.03 for the Suwannee River population. T he CJS model estimates might be biased due to their model assumptions, mainly that all individuals have the same capture probability (i.e. the population might be sampled non randomly in known hold ing areas). Under this assumption, it might not be reasonable to assume the marked population is representative of the unmarked population. Regardless, Suwannee River Gulf sturgeon population abundance estimates from this study and the CJS model used by US GS are both much higher than abundance estimates in any other riverine population. Although detailed model results are not currently available for
101 other rivers, USFWS estimated a much lower population in the Apalachicola River as of 2004 (USFWS: 350 indivi duals; this study: 1,625 individuals). Parameter estimates for Gulf sturgeon from ASMR analysis are more reliable due to the incorporation of age structure and age assignment uncertainty prior information from previous studies to inform objective paramete rs (e.g. natural mortality), and consideration of both the unmarked and marked population in cohort abundance estimates. The spikes in capture probability occur in years when a greater number of tags were implanted in Gulf sturgeon relative to previous ye ars (Figure 4 1 ). However, capture probability also coincides with the patterns in age 1 recruitment and abundance of Gulf sturgeon age 2 and older. The 1992 peak in capture probability coincided with the peak in age 1 recruitment in 1991 (Figure 4 5) and peak in Gulf sturgeon age 2 and older abundance in 1992 (Figure 4 3). The coincidence of patterns in these parameters indicates a potential change in sampling design within the Suwannee River. Assuming a constant mortality rate, capture probability and abu ndance estimates would be expected to be high with a large cohort of newly marked Gulf sturgeon assigned young ages. The ASMR model, ignorant of sampling design, would assume the spike in captured Gulf sturgeon was related to a spike in recruitment, which in the case of Gulf sturgeon is likely false. Although the bifurcation of the abundance trajectories in the Apalachicola River with different random age assignments is an interesting finding relevant for management, I did not see this pattern when using a Po isson probability di stribution. More simulation is needed to specifically identify when managers should be wary of bias in their parameter estimates. The shapes of the Poisson and negative binomial
102 distributions explain the mechanics of observed variat ion, but not why do some age assignments lead to decreasing populatio n trajectories, and some do not. The answer to this question could only be confirmed through detailed simulation, which will be an interesting future venture. Possibilities to be tested w ould be a potential bias towards assigning Gulf sturgeon to younger ages at the inflexion point of the cumulative normal probability distribution. If a large number of Gulf sturgeon are randomly assigned to the same, young cohort, it is possible they could still be invulnerable to the capture gear. As they cross the inflexion point of the vulnerability schedule, they could all become vulnerable at once. This could lead to the high spikes in capture probability for age assignment iterations estimating a down ward population trajectory. These issues could to humpback chub, Coggins (2008) observed this pulsed effort sampling associated with uncertainty in abundance, but did not observe bifurcation in abundance trajectories with iterations of age assignment. Another possible issue was an uninformat ive prior on natural mortality that was set at 0.08 based on the lowest estimate of pr evious studies (Flowers 2008) However, this study estimates a much lower natural mortality rate, but in a few iterations the potentially uninformative prior may be forcing natural morta lity higher than it should be (0.04 0.07 Figure 4 7 ) leading to lower population estimates in those cases. Future simulation is necessary to confirm the reasoning behind the bifurcation in abundance estimates of Gulf sturgeon in the Apalachicola River an d corresponding uncertainty in natural mortality and capture probability parameters. My assessment provides more certain population parameter estimates for Gulf sturgeon that managers can apply with greater confidence. Gulf sturgeon managers
103 now have the results from multiple approaches to estimate population abundance in the Suwannee River. The Suwannee River certainly supports the largest population of Gulf sturgeon as expected, and the ASMR estimates of abundance have less uncertain ty than mark recaptu re methods alone. Additionally, the Apalachicol a population of Gulf sturgeon had increasing abundance over time. In their population viability analysis for the Suwannee River, Pine et al. (2001) est imated population growth rate as 1.05, indicating population growth over time. This supports my results that the Suwannee River abundance is stable and increasing over the time period. This is encouraging to managers interested in the recovery of this species. With information on carrying capac ity for each river drainage, combined with quantifiable, defined recovery criteria, these Gulf sturgeon populations could be on the road to recovery. However, Gulf sturgeon managers should be mindful of the uncertainties in abundance estimates associated w ith age assignment when tagg ing effort is highly variable. To better understand the status of Gulf sturgeon in the Pearl and Pascagoula Rivers where CPUE is low and telemetry predicts very high mortality, more tags need to be deployed annually. This could be accomplished through re allocation of resources towards the areas that need it, and away from drainages like the Suwannee where so much information is known and the population is doing well.
104 Table 4 1. Number of initial marks and recaptures of Gulf sturgeon with internal and external passive tags in each river from 1977 2012. Year Suwannee Apalachicola Marks Recaptures Marks Recaptures 1977 0 0 1 0 1978 0 0 3 0 1979 0 0 1 0 1980 0 0 1 0 1981 0 0 0 0 1982 3 0 30 0 198 3 0 0 62 13 1984 35 0 27 19 1985 0 0 47 40 1986 320 1 19 31 1987 261 31 27 24 1988 387 74 28 31 1989 403 109 19 15 1990 780 207 61 11 1991 460 190 12 9 1992 1687 182 8 2 1993 719 243 50 11 1994 512 202 0 0 1995 554 327 1 0 4 1996 142 82 0 0 1997 244 170 1 0 1998 454 233 73 14 1999 369 221 152 20 2000 95 75 2 1 2001 157 80 42 11 2002 136 68 40 22 2003 34 15 25 7 2004 0 0 140 41 2005 34 6 66 18 2006 304 98 162 47 2007 204 63 17 8 2008 0 0 0 0 2009 0 0 0 0 2010 15 5 172 19 2011 19 1 116 24 2012 0 0 28 13 Total 83 28 2683 1 442 4 55
105 Table 4 2. Total number of Gulf sturgeon and percentage of Gulf sturgeon tagged in each river that emigrated outside of the river in which they were tagged. River Emigrants % Emigrants Suwannee 5 0.1 Apalachicola 20 4.1 Note: Between 1977 and 2012.
106 Table 4 3. Median estimates of Gulf sturgeon natural mortality from the MCMC posterior distribution with constructed upper (UCL) and lower (LCL) 95% confidence intervals. Suwannee Apalachicola Estimate 0.031 0.032 LCL 0.024 0.026 UCL 0.039 0.039
107 Figure 4 1 The number of initial marks and recaptures of Gulf sturgeon at age over time Numbers are expressed by bubble size. A) Suwannee River. B) Apalachicola River. A B
108 Figure 4 2 Von Bertalanffy growth curve fit to limited age at length data shows a high variation in ages as length increases.
109 Figure 4 3 Abundan ce estimates i n numbers for age 2+ Gulf sturgeon populations over time. Median estimates and 95% confidence intervals from the MCMC posterior distribution, assuming a negative bi nomial probability distribution incorporating 2 ,000 iterations of age assignment. A) Suwannee population. B) Apalachicola population. A B
110 Figure 4 4 Maximum likelihood esti mates of abundance trajectories of Apalachicola River Gulf sturgeon compared with MCMC estimates. Each thin line represents a different iterate of 2,000 iterations of initial age assignme nt.
111 Figure 4 5. Numbers of age 1 Gulf sturgeon recruits over time. Median estimates and 95% confidence intervals from the MCMC posterior distribution, assuming a negative binomial probability distribution, incorporating 2,000 iterations of age assignm ent. A) Suwannee population. B) Apalachicola population. A B
112 Figure 4 6 Capture probability of Gulf sturgeon over time. The median and 95% confidence intervals were calculated from a distribution of maximum likelihood estimates from 2,000 iterations of i nitial age assignment assuming a negative binomial probability distribution A) Suwannee River. B) Apalachicola River. A B
113 Figure 4 7 Divergence in maximum likelihood parameter estimates between sample iterates of age assignment for the Apalachicola River Gul f sturgeon population. Estimates were obtained assuming a negative binomial probability distribution A) Capture probability B) N atural mortality A B
114 Figure 4 8 Comparison of P oisson and negative binomial probability distributions for the number of predicted Gulf st urgeon recaptures in the year 2004 The year 2004 was chosen as one of the years where ASMR analysis under the negative binomial assumption estimated bifurcated abundance esti mates in the Apalachicola River.
115 Figure 4 9 Suwannee River Gulf sturgeon historic mark recapture data and population parameters estimated from CJS mark recapture model. A) Gulf sturgeon c aptures. B) Estimated captured probability. C) E stimated abundance over time [Adapted from M. Randall, USGS, unpublis hed data] A B C
116 CHAPTER 5 CO NCLUSIONS AND RECOMMENDATIONS This study confirmed that the standardized acoustic telemetry survey will be successful at estimating unbiased survival, detection probability, and movement rates of Gulf sturgeon after the full five year monitoring period fo r the majority of rivers being assessed Detection probability across all states and both migration seasons was much greater than the capture probability associated with physical recaptures in previous studies, indicating yet another success of the five ye ar survey However, a limitation of the telemetry survey was unavoidable imprecision around demographic parameters associated with a relatively short monitoring program for a long lived species like Gulf sturgeon, and a relatively low number of acoustic ta gs due to limited resources The simulations done by Zehfuss (2000) support the relatively short time frame because detection probability is so high. However, uncertainties associated with lim ited number of tags are more difficult to resolve. It is important to remember the demographic rates for G ulf sturgeon estimated through the multi state model are based only on the tagged population, a potentially small sample of the entire population. In Gulf sturgeon populations such as that in the Choctawhatchee River I can hope the high number of acoustically tagged individuals is an adequate representation of the movement and mortality for the entire riverine population. Although the goal of acoustica lly tagging 40 individuals in the Suwannee River was met, the large number of individuals in that population limits my ability to be certain in estimates and management recommendations solely from telemetry. Movement rates between regions of the Gulf hav e important implications for management of the Gulf sturgeon Given natal homing and minimal mixing of Gulf
117 sturgeon in the riverine populations, it is reasonable to c onclude a low percentage of the Gulf sturgeon from eastern rivers move into rivers in the western Gulf. High fidelity rates are also important for designating stock structure and could be useful to identify and manage threats to Gulf sturgeon at a riverine scale Treating western Gulf populations separately from central or eastern riverine pop ulations could create independent pools of resources for management, allocating funding appropriately to areas in need. The methods used to estimate Gulf wide movement rates of Gulf sturgeon simultaneously estimated area specific survival and detection p robability. In the eastern and central Gulf sturgeon populations (Suwannee, Ochlockonee, Apalachicola, and Choctawhatchee Rivers, as well as river in Escambia Bay), survival rates appear high and relatively precise given the larger sample size of tagged Gu lf sturgeon, but precision and accuracy of estimates are expected to increase with five full years of monitoring. Survival rates are most uncertain in the western Gulf (Pearl and Pascagoula Rivers) due to the low number of tagged Gulf sturgeon, and are exp ected to remain uncertain through the five year survey if tagging rates remain the same. This is especially concerning due to high potential for anthropogenic impacts from habitat changes, pollution, and bycatch that remain un quantified. The number of aco ustic transmitters deployed in Gulf sturgeon in the Pearl and Pascagoula Rivers is far below the goal for the study, not due to lack of effort, but due to potentially lower catchability and sampling restrictions preventing Gulf sturgeon researchers from sa mpling at the most efficient periods. Regardless, unless the number of tagged Gulf sturgeon in these rivers increases, results from the five year survey will fail to provide data needed to
118 estimate unbiased, precise survival rates for the Pearl and Pascago ula riverine populations. The ASMR analysis results provide an optimistic view of Gulf sturgeon populations in the eastern Gulf. The Suwannee River population appears high and stable, with low uncertainty associated with initial age assignment. The Apala chicola River has a high cumulative tagged population, but with more pulsed effort over time. Using a Poisson probability distribution, ASMR consistently estimates an increasing population trajectory with relatively high precision when incorporating uncert ainty in age assignment. However, it is import ant to identify the possibility of estimating a declining population trajectory when using a negative binomial probability distribution. Managers would be remiss not to properly test possible probability distr ibutions and understand uncertainties underlying the likelihood structures. Uncertainty associated with initial age assignment in the Apalachicola River calls for a more constant mark recapture program Funds allocated into smaller amount s over a cons tant and longer period of time could help provide unbiased, time specific estimates of abundance. Even if it means reaching a high cumulative tagged population at a slower rate, managers should further explore the implications of a pulsed effort tagging program Managers should be aware that a shared downfall of the short term acoustic telemetry and constant, low monitoring effort for PIT tagging is the potential inability to properly identify population impacts. As the cumulative population of PIT tags increa ses over time, it will become easier to identify recent population impacts with more certainty. The short term battery life of acoustic transmitters and expenses associated with maintaining an acoustic array are major limitations to acquiring a large numbe r of
119 individuals in the study over a long period of time. To detect changes in a population from a major mortality event, the telemetry array and an acoustically tagged population must be at The small number of Gulf stu rgeon with acoustic tags in the Pearl River, for example, was not sufficient to determine the impacts of the major paper mill byproduct spill during the first year of monitoring. On the other hand, a large cumulative number of individuals in a tagged popul ation over time, analyzed using ASMR, could be the better sampling design for identifying significant population change. High fidelity, differential mortality and uncertainty associated with variable quality mark recapture data, combined with genetic stu dies (Stabile et al. 1996) provide excellent information to identify stock structure for the Gulf sturgeon ( Grimes et al. 1987 ; Coyle 1998; Begg and Waldman 1999 ) Stock structure would be the first step towards recognizing distinct population segments (DPS) of the Gulf sturgeon that could ultimately lead to d e listing of well understood, recovered Gulf sturgeon populations is dependent on the definition of quantifiable, redefined recovery criteria Estimated abundance compared to riverine carrying capacity could also be considered when defining a recovered riv erine population of Gulf sturgeon Stock reduction analysis (Ahrens and Pine, in preparation) has estimated carrying capacities similar to or lower than current estimated population size for some rivers, an indication that abundance of those Gulf sturgeon populations may be approaching carrying capacity However, the carrying capacities would need to incorporate habitat changes post exp loitation, so current abundance of Gulf sturgeon could not be compared to pre exploitation abundance. Population viability analysis would also be necessary to assess if Gulf
120 sturgeon populations would be able to persist given habitat changes and associated carrying capacities. If some Gulf sturgeon populations meet these redefined, quantifiable recovery criteria, resources sho uld be allocated towards the areas where more mark recapture information is needed. Gulf sturgeon as a species, managed as one population, will not be able to be de listed by the goal year 2023 if parameter estimates in the western Gulf remain uncertain an d hint to a declining population. On the other hand, if a DPS were to be designated, riverine populations could be considered for delisting and a renewed fishery established in the eastern Gulf to attract associated economic benefits. Otherwise, the c ontin ued management of Gulf sturgeon as a species across its range may continue to inefficiently allocate resources to the east instead of the western Gulf, where the population dynamics are not well understood. The most important data need for Gulf sturgeon is increasing the cumulative tagged population constantly over time in the Pearl, Pascagoula, Escambia, Blackwater, Yellow, Choctawhatchee, Ochlockonee, and even Apalachicola Rivers to estimate population parameters with precision and make better inference s in stock status. An immediate research need is the simulation of tagging effort patterns over time, the size of the cumulative tagged population compared to the total population size, and the inflexion points in the curves for vulnerability and cumulativ e probability of age given length to better understand the impacts of age assignment uncertainty. This simulation can be used to determine the number of years of tagging effort necessary in the Apalachicola River to ensure certainty in their increasing pop ulation trajectory and low mortality rate. Although detailed, it is important to identify the reasons behind bifurcation in the Apalachicola River population trajectories under the negative binomial
121 assumption to understand the certainty of increasing popu lation trajectory. Simulation is also needed to understand the accuracy of ASMR with limited data and age assignment uncertainty With the causes of uncertainty quantified, my recommendations to managers would be more specific and applicable.
122 LIST OF REFERENCES Akaike, H. 1973. Maximum likelihood identification of Gaussian autoregressive moving average models. Biometrika 60(2):255 265. Arnason, A. N., and K. H. Mills. 1981. Bias and loss of precision due to tag loss in Jolly Seber estimates for mark recapture experiments. Canadian Journal of Fisheries and Aquatic Sciences 38(9):1077 1095. Baremore, I. E., and J. D. Rosati. 2011. Gulf sturgeon standardized abundance and mortality study: year one report. NMFS SEFSC. Baremore, I. E. and J. D. Rosati. 2012. Gulf sturgeon standardized abundance and mortality study: year two report. NMFS SEFSC. Begg, G., D. Cameron, and W. Sawynok. 1997. Movements and stock structure of school mackerel (Scomberomorus queenslandicus) and spotted macker el (S munroi) in Australian east coast waters. Marine and Freshwater Research 48(4):295 301. Begg, G., and J. Waldman. 1999. An holistic approach to fish stock identification. Fisheries Research 43(1 3):35 44. Berg, J. J. 2004. Population assessment of t he Gulf of Mexico sturgeon in the Yellow River, Florida. University of Florida, Gainesville, Florida. Berg, J. J., M. S. Allen, and K. J. Sulak. 2007. Population assessment of the Gulf of Mexico sturgeon in the Yellow River, Florida. American Fisheries So ciety Symposium 56:365 379. Bridger, C., and R. Booth. 2003. The effects of biotelemetry transmitter presence and attachment procedures on fish physiology and behavior. Reviews in Fisheries Science 11(1):13 34. Brodziak, J., J. Ianelli, K. Lorenzen, and R. D. J. Method. 2011. Estimating natural mortality in stock assessment applications. Alaska Fisheries Science Center, Seattle, WA. Brown, B. E., G. H. Darcy, and W. Overholtz. 1987. Stock assessment/stock identification: an interactive process. U.S. Depa rtment of Commerce, Washington, D.C. Brownie, C., J. E. Hines, J. D. Nichols, K. H. Pollock, and J. B. Hestbeck. 1993. Capture recapture studies for multiple strata including non Markovian transitions. Biometrics 49(4):1173 1187.
123 Carr, S. M., A. J. Snell en, K. A. Howse, and J. S. Wroblewski. 1995. Mitochondrial DNA sequence variation and genetic stock structure of Atlantic cod ( Gadus morhua ) from bay and offshore locations in the Newfoundland continental shelf. Molecular Ecology 4(1):79 88. Carvalho, G. R., and L. Hauser. 1994. Molecular genetics and the stock concept in fisheries. Reviews in Fish Biology and Fisheries 4(3):326 350. Casselman, J. M., J. J. Collins, E. J. Crossman, P. E. Ihssen, and G. R. Spangler. 1981. Lake whitefish ( Coregonus clupeafo rmis ) stocks of the Ontario waters of Lake Huron. Canadian Journal of Fisheries and Aquatic Sciences 38(12):1772 1789. Chapman, F. A., C. S. Hartless, and S. H. Carr. 1997. Population size estimates of sturgeon in the Suwannee River, Florida, USA. Gulf of Mexico Science 15:88 91. Clark, W. 1999. Effects of an erroneous natural mortality rate on a simple age structured stock assessment. Canadian Journal of Fisheries and Aquatic Sciences 56(10):1721 1731. Coggins, L., and coauthors. 2006a. Abundance trends and status of the little Colorado River population of humpback chub. North American Journal of Fisheries Management 26(1):233 245. Coggins, L. G., W. E. Pine, C. J. Walters, and S. J. D. Martell. 2006b. Age structured mark recapture analysis: A virtual p opulation analysis based model for analyzing age structured capture recapture data. North American Journal of Fisheries Management 26(1):201 205. Coggins, L. G. J. 2007. Abundance Trends and Status of the Little Colorado River Population of Humpback Chub: An Update Considering 1989 2006 Data. U.S. Geological Survey. Coggins, L. G. J. 2008. Active adaptive management for native fish conservation in the Grand Canyon: implementation and evaluation. University of Florida, Gainesville, Florida. Coggins, L. G. and C. J. Walters. 2009. Abundance trends and status of the Little Colorado River population of humpback chub: an update considering data from 1989 2008. Conn, P., W. Kendall, and M. Samuel. 2004. A general model for the analysis of mark resight, mark r ecapture, and band recovery data under tag loss. Biometrics 60(4):900 909.
124 Coyle, T. 1998. Stock identification and fisheries management: the importance of using several methods in a stock identification study. Pages 173 182 in D. A. Hancock, editor. Tak ing stock: defining and managing shared resources. Australian Society for Fish Biology, Sydney. Cushing, D. H. 1968. Fisheries biology: a study in population dynamics. University of Wisconsin Press, Madison. Daniel, A., B. Hicks, N. Ling, and B. David. 2 009. Acoustic and radio transmitter retention in common carp (Cyprinus carpio) in New Zealand. Marine and Freshwater Research 60(4):328 333. Dobson, A., J. Rodriguez, and W. Roberts. 2001. Synoptic tinkering: Integrating strategies for large scale conserv ation. Ecological Applications 11(4):1019 1026. Ebbin, S. 1996. The stock concept: Constructing tools for Pacific salmon management. Coastal Management 24(4):355 364. Louisiana Department of Wildlife and Fisheries. 2011. Investigation of a fish and mollu sk kill in the lower Pearl River, Louisiana and Mississippi. Louisiana Department of Wildlife and Fisheries, Baton Rouge. Flowers, H., and coauthors. 2009. Spawning Site Selection and Potential Implications of Modified Flow Regimes on Viability of Gulf St urgeon Populations. Transactions of the American Fisheries Society 138(6):1266 1284. Flowers, H. J. 2008. Age structured population model for evaluating Gulf sturgeon recovery on the Apalachicola River, Florida. University of Florida, Gainesville, Florida Fournier, D., and coauthors. 2012. AD Model Builder: using automatic differentiation for statistical inference of highly parameterized complex nonlinear models. Optimization Methods & Software 27(2):233 249. Fox, D., J. Hightower, and F. Paruka. 2000. Gulf sturgeon spawning migration and habitat in the Choctawhatchee River system, Alabama Florida. Transactions of the American Fisheries Society 129(3):811 826. Fox, D. A., J. E. Hightower, and F. M. Parauka. 2002. Estuarine and nearshore marine habitat u se by Gulf sturgeon from the Choctawhatchee River system, Florida. American Fisheries Society Symposium 00:19 34. Gelman, A. B., J. S. Carlin, H. S. Stern, and D. B. Rubin. 2000. Bayesian data analysis. Chapman and Hall, New York.
125 Grimes, C. B., A. G. Jo hnson, and W. A. J. Fable. 1987. Delineation of king mackerel ( Scomberomorus cavalla ) stocks along the US east coast and in the Gulf of Mexico. Pages 186 187 in H. E. Kumpf, R. N. Vaught, C. B. Grimes, A. G. Johnson, and E. L. Nakamura, editors. Stock Iden tification Workshop. US Government Printing Office, Panama City Beach, FL. Grothues, T. M. 2009. A review of acoustic telemetry technology and a perspective on its diversification relative to coastal tracking arrays. J. L. Nielsen, and coeditors, editors. Tagging and tracking of marine animals with electronic devices, reviews: methods and technologies in fish biology and fisheries., volume 9. Springer, Netherlands. Gulland, J. A. 1983. Fish stock assessment: a manual of basic methods, volume 1. John Wiley and Sons, New York. Hestbeck, J. B., J. D. Nichols, and R. A. Malecki. 1991. Estimates of movement and site fidelity using mark resight data of wintering Canada geese. Ecology 72(2):523 533. Heupel, M., J. Semmens, and A. Hobday. 2006. Automated acousti c tracking of aquatic animals: scales, design and deployment of listening station arrays. Marine and Freshwater Research 57(1):1 13. Hewitt, D., and J. Hoenig. 2005. Comparison of two approaches for estimating natural mortality based on longevity. Fishery Bulletin 103(2):433 437. Hightower, J., J. Jackson, and K. Pollock. 2001. Use of telemetry methods to estimate natural and fishing mortality of striped bass in Lake Gaston, North Carolina. Transactions of the American Fisheries Society 130(4):557 567. H ilborn, R. 1990. Determination of fish movement patterns from tag recoveries using maximum likelihood estimators. Canadian Journal of Fisheries and Aquatic Sciences 47(3):635 643. Hilborn, R., and C. J. Walters. 1992. Quantiative Fisheries Stock Assessmen t: Choice, Dynamics, and Uncertainty. Chapman and Hall, New York. Huff, J. A. 1975. Life history of Gulf of Mexico sturgeon, Acipenser oxyrinchus desotoi in Suwannee River, Florida. Florida Marine Research Publications 16. Ihssen, P. E., and coauthors. 1981. Stock identification materials and methods. Canadian Journal of Fisheries and Aquatic Sciences 38(12):1838 1855. Irvine, R. L., D. C. Schmidt, and L. R. Hildebrand. 2007. Population status of white sturgeon in the lower Columbia River within Canad a. Transactions of the American Fisheries Society 136(6):1472 1479.
126 Jensen, A. 1996. Beverton and Holt life history invariants result from optimal trade off of reproduction and survival. Canadian Journal of Fisheries and Aquatic Sciences 53(4):820 822. J ohnson, A. G., W. A. Fable, C. B. Grimes, L. Trent, and J. V. Perez. 1994. Evidence for distinct stocks of king mackerel, Scomberomorus cavalla in the Gulf of Mexico. Fishery Bulletin 92(1):91 101. Kahnle, A. W., and coauthors. 1998. Stock status of Atla ntic sturgeon of Atlantic coastal estuaries. Atlantic States Marine Fisheries Commission. Kendall, W., J. Nichols, and J. Hines. 1997. Estimating temporary emigration using capture recapture data with Pollock's robust design. Ecology 78(2):563 578. Laure tta, M., E. Camp, W. Pine, and T. Frazer. 2013. Catchability model selection for estimating the composition of fishes and invertebrates within dynamic aquatic ecosystems. Canadian Journal of Fisheries and Aquatic Sciences 70(3):381 392. Lebreton, J. D., K P. Burnham, J. Clobert, and D. R. Anderson. 1992. Modeling survival and testing biological hypotheses using marked animals a unified approach with case studies. Ecological Monographs 62(1):67 118. Lorenzen, K. 2000. Allometry of natural mortality as a basis for assessing optimal release size in fish stocking programmes. Canadian Journal of Fisheries and Aquatic Sciences 57(12):2374 2381. Lowe, W. 2002. Landscape scale spatial population dynamics in human impacted stream systems. Environmental Manageme nt 30(2):225 233. McMichael, G., and coauthors. 2010. The Juvenile Salmon Acoustic Telemetry System: A New Tool. Fisheries 35(1):9 22. Melnychuk, M. 2009. Estimation of survival and detection probabilities for multiple tagged salmon stocks with nested mi gration routes, using a large scale telemetry array. Marine and Freshwater Research 60(12):1231 1243. Morrow, J. V., J. P. Kirk, J. K. Killgore, H. Rogillio, and C. Knight. 1998. Status and recovery potential of Gulf sturgeon in the Pearl River system, Lo uisiana Mississippi. North American Journal of Fisheries Management 18:798 808. Nichols, J. D., and K. H. Pollock. 1983. Estimation methodology in contemporary small mammal capture recapture studies. Journal of Mammalogy 64(2):253 260. NMFS and USFWS. 20 10. Gulf sturgeon sampling protocol and monitoring survey. 2pp.
127 NOAA. 2012 Natural Resource Damage Assessment April 2012 Status Update for the Deepwater Horizon Oil Spill. Pages www.gulfspillrestora tion.noaa.gov in NOAA, NMFS, and SERO. 2012. Endangered Species Act Section 7 Consultation Biological Opinion. National Oceanic and Atmospheric Administration, National Marine Fisheries Service, Southeast Regional Office, Protected Resources Division and Sustainable Fisheries Division. Parauka, F., M. Duncan, and P. Lang. 2011. Winter coastal movement of Gulf of Mexico sturgeon throughout northwest Florida and southeast Alabama. Journal of Applied Ichthyology 27(2):343 350. Pauly, D. 1980. On the i nterrelationships between natural mortality, growth parameters, and mean environmental temperature in 175 fish stocks. Journal Du Conseil 39(2):175 192. Pine, W., M. Allen, and V. Dreitz. 2001. Population viability of the gulf of Mexico sturgeon: Inferenc es from capture recapture and age structured models. Transactions of the American Fisheries Society 130(6):1164 1174. Pine, W. E., and S. J. D. Martell. 2009. Status of Gulf sturgeon Acipenser oxyrinchus desotoi in the Gulf of Mexico: A document prepared for review, discussion, and research planning. Gulf sturgeon annual working group meeting, Cedar Key, FL. Pine, W. E., K. H. Pollock, J. E. Hightower, T. J. Kwak, and J. A. Rice. 2003. A review of tagging methods for estimating fish population size and co mponents of mortality. Fisheries 28(10):10 23. Pollock, K. H., J. D. Nichols, C. Brownie, and J. E. Hines. 1990. Statistical inference for capture recapture experiments. Wildlife Monographs (107):1 97. Pollock, K. H., and coauthors. 2002. Large scale wil dlife monitoring studies: statistical methods for design and analysis. Environmetrics 13:105 119. Rogillio, H. E., and coauthors. 2001. Status, movement, and habitat use of Gulf sturgeon in the Lake Pontchartrain basin, Louisiana. Louisiana Department of Wildlife and Fisheries and National Fish and Wildlife Foundation, Baton Rouge. Ross, S. T., R. J. Heise, M. A. Dugo, and W. T. Slack. 2001. Movement and habitat use of the Gulf sturgeon ( Acipenser oxyrinchus desotoi ) in the Pascagoula drainage of Mississi ppi: year V. Department of Biological Sciences, University of Southern Mississippi, and Mississippi Museum of Natural Science. Ross, S., and coauthors. 2009. Estuarine and Coastal Habitat Use of Gulf Sturgeon (Acipenser oxyrinchus desotoi) in the North Ce ntral Gulf of Mexico. Estuaries and Coasts 32(2):360 374.
128 Schaub, M., O. Gimenez, B. Schmidt, and R. Pradel. 2004. Estimating survival and temporary emigration in the multistate capture recapture framework. Ecology 85(8):2107 2113. Skillman, R. A. 1989. Stock identification and billfish management. Pages 207 214 in R. H. Stroud, editor Planning the Future of Billfishes, Research and Management in the 90s and Beyond. National Coalition for Marine Conservation Inc., Savannah, GA. Stabile, J., J. Waldman, F Parauka, and I. Wirgin. 1996. Stock structure and homing fidelity in gulf of Mexico sturgeon (Acipenser oxyrinchus desotoi) based on restriction fragment length polymorphism and sequence analyses of mitochondrial DNA. Genetics 144(2):767 775. Stobo, W. T., J. S. Scott, and J. J. Hunt. 1975. Movement of herring tagged in the Bay of Fundy. International Commission for the Northwest Atlantic Fisheries Research Document. Sulak, K. J., and J. P. Clugston. 1999. Recent advances in life history of Gulf of Mexi co sturgeon, Acipenser oxyrinchus desotoi, in the Suwannee river, Florida, USA: a synopsis. Journal of Applied Ichthyology Zeitschrift Fur Angewandte Ichthyologie 15(4 5):116 128. USFWS. 1995. Gulf sturgeon recovery plan. US Fish and Wildlife Service and Gulf States Marine Fisheries Commission. USFWS. 2007. Panama City Fisheries Resources Office 2007 Annual Report. USFWS. 2009. Panama City Fisheries Resources Office 2009 Annual Re port. USFWS and NMFS. 2009. Gulf sturgeon five year review: summary and evaluation. Walters, C., B. van Poorten, and L. Coggins. 2012. Bioenergetics and Population Dynamics of Flannelmouth Sucker and Bluehead Sucker in Grand Canyon as Evidenced by Tag Re capture Observations. Transactions of the American Fisheries Society 141(1):158 173. Welch, D., S. Batten, and B. Ward. 2007. Growth, survival, and tag retention of steelhead trout (O mykiss) surgically implanted with dummy acoustic tags. Hydrobiologia 58 2:289 299. Welch, D., and coauthors. 2009. Freshwater and marine migration and survival of endangered Cultus Lake sockeye salmon (Oncorhynchus nerka) smolts using POST, a large scale acoustic telemetry array. Canadian Journal of Fisheries and Aquatic Scie nces 66(5):736 750.
129 White, G., and K. Burnham. 1999. Program MARK: survival estimation from populations of marked animals. Bird Study 46:120 139. Williams, B. K., J. D. Nichols, and M. J. Conroy. 2001. Analysis and management of animal populations. Acade mic Press, London. Wooley, C. M., and E. J. Crateau. 1985. Movement, microhabitat, exploitation, and management of Gulf of Mexico sturgeon, Apalachicola River, Florida. North American Journal of Fisheries Management 5:590 605. Zehfuss, K. P., J. E. Hight ower, and K. H. Pollock. 1999. Abundance of Gulf sturgeon in the Apalachicola River, Florida. Transactions of the American Fisheries Society 128(1):130 143. Zehfuss, K. P. 2000. The status, movement, habitat preference, and monitoring of Gulf sturgeon in several Florida rivers. North Carolina State University, Raleigh, North Carolina.
130 BIOGRAPHICAL SKETCH Merrill Rudd was born in Livingston, New Jersey in 1989 to Howard and Eileen Rudd. She was raised in Short Hills, New Jersey, graduating from Millb urn Senior High School in 2007. Despite an interest in marine eco systems, Merrill traveled inland to attend Washington University in St. Louis. As an undergraduate she took every opportunity to conduct field research, ranging from tree distributions in Mis souri, urban coastal ecology in the New York/New Jersey area, to small scale fisheries dynamics in Panama. Merrill was introduced to quantitative fisheries science at the NOAA RTR Marine Resources Population Dynamics undergraduate workshop, which combined her interest in marine ecology, math, and eating seafood. In 2011, Merrill earned her B.A. in Environmental Studies with concentrations in ecology and biology at Washing ton University. Pursuing her developed interest in quantitative fisheries science, Merr ill received her M.S. from the University of Florida in the summer of 2013. She will be in fall 2013 to pursue a Ph.D. under Drs. Trevor Branch and Ray Hilborn, with a fellowsh ip from the IGERT Program on Ocean Change.