THE IMPACT OF TIMBER HARVEST ON WILDLIFE DISTRIBUTION PATTERNS AND POPULATION VITAL RATES: DOES STRUCTURAL RETENTION AMELIORATE THE NEGATIVE EFFECTS OF CLEARCUTTING? By Clint R.V. Otto A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Fisheries and Wildlife Ecology, Evolutionary Biology, and Behavior 2012 ABSTRACT THE IMPACT OF TIMBER HARVEST ON WILDLIFE DISTRIBUTION PATTERNS AND POPULATION VITAL RATES: DOES STRUCTURAL RETENTION AMELIORATE THE NEGATIVE EFFECTS OF CLEARCUTTING? By Clint R.V. Otto Anthropogenic habitat degradation is a primary threat to global biodiversity. The overarching theme of my dissertation is one of conservation-based research for reducing the impacts of anthropogenic disturbances, specifically timber harvesting, on forest wildlife. Although the primary objective of my dissertation is to assess the effectiveness of structural retention (i.e., canopy green trees and coarse woody debris) for conserving wildlife in harvested forests, it also provides valuable insight to other areas of ecology and wildlife management. Throughout my dissertation I highlight the need to use design- and model-based approaches for minimizing bias associated with the limited detectability of wildlife. Doing so allowed me to achieve stronger inference when determining how wildlife interacts with their environment. For my first two chapters I use species occupancy models and multiple sampling techniques to estimate detection probabilities of forest-floor wildlife. These chapters demonstrate the need to incorporate species detectability when comparing the effectiveness of different trapping methodologies. Furthermore, these studies highlight the utility of power analyses for exploring study design tradeoffs for research and monitoring programs. I used results from these chapters to develop a strong sampling design for Chapter 5. In Chapter 3 I lead a detailed investigation into the performance of wildlife occupancy models when model assumptions were openly violated. Analyses from Chapter 3 revealed general sensitivity of estimates from single-season occupancy models to violations of closure. This chapter highlights the importance of addressing the     population “closure” assumption and the impact of non-random organism movements in wildlife occupancy studies. I provide multiple solutions for minimizing bias associated with non-random changes in occupancy within a field season. For my final two chapters I assess the effectiveness of structural retention for conserving wildlife in harvested forests. Both studies represent an empirical evaluation of the Michigan Department of Natural Resources (MDNR) structural retention guidelines in harvested aspen stands. Results from Chapters 4 and 5 suggest that broad-scale conservation goals for forest songbirds and terrestrial salamanders will not be accomplished by simply retaining structure within individual harvest units. Rather, successful conservation efforts for these species may also require conservation of late-successional forests as part of managed landscapes. Future management objectives in aspen forests should reflect the value of clearcutting to early-successional and generalist bird species while recognizing that green-tree and coarse woody debris (CWD) retention, as observed here, did little to promote site occupancy of interior forest bird and terrestrial salamanders. However, if the management goal is to reduce mortality of local salamander populations following timber harvest, then my research suggests that implementing structural retention prescriptions is a viable management option. Results from Chapter 5 demonstrate that organismal pattern (occupancy) and process (survival) can function independently in response to habitat change, and thus emphasizes the importance joint-evaluation of these state variables when evaluating mitigation techniques presumed beneficial to wildlife. ACKNOWLEDGEMENTS Support for my dissertation was provided by the MDNR-Wildlife Division with funds from the federal Pittman-Robertson Wildlife Restoration Act grant administered by the United States Fish and Wildlife Service (W-147-R: Michigan’s Statewide Wildlife Research and Restoration Program). I also received funding support from the Rocky Mountain Goat Foundation, the Rachana Rajendra Conservation Fellowship, the Michigan Society of Herpetologists, the Michigan State University (MSU) Ecology, Evolutionary Biology and Behavior Program, the MSU Graduate School, and the MSU Department of Fisheries and Wildlife. I greatly appreciate the mentorship and friendship of my advisor Gary Roloff, who always treated me as a colleague and provided me with the utmost intellectual freedom. Thanks to my committee members Larissa Bailey, Scott Winterstein, and David MacFarlane for their support throughout my dissertation. In particular, I thank Larissa Bailey for her steadfast guidance throughout my professional career. I also want to extend my gratitude to our Departmental support staff, particularly Marcia Baar, for always being there to lend a helping hand. I am indebted to the service of numerous technicians, MDNR employees, and volunteers for their involvement in my research. I appreciate the exceptional service provided by David Burt, Andrew Coleman, Jonathan Lautenbach, Joseph Lautenbach, Darin McNeil, Rachelle Sterling, Rachael Thames, and Shayna Wieferich. I wish you all the best on your future personal and professional endeavors. A special thanks to Mike Donovan, Kerry Fitzpatrick, and Dan Moran (MDNR) for their efforts throughout my dissertation. Working with these individuals and the MDNR has been a very rewarding and enlightening experience. iv    I would also like to thank my friends and family. Whether in Michigan or abroad, past or present, I appreciate your support. I could have never succeeded without you. In particular, I want to thank the Greenways, Grandma Marcia, Emily Johnston, Andy Flies, Dan Linden, and Leslie Curren for helping with care of my children whenever Tammy and I needed a hand. I thank Gerard and Kim Fuehrer (i.e., Papa and Nana) for their support throughout my dissertation and for nurturing my children during a lengthy 2011 field season. To my parents Dick and Pat (i.e., Bopu and Grandma): the value system in which you have instilled has served me well throughout life. Thank you for your love and guidance. I thank my daughters Coralyn and Aurelia (i.e., the Bumps) for giving me a new life perspective: Your Mother and I look forward to a lifetime of adventure, love, and learning with you. I am forever indebted to the undying support and sacrifice of my wife and friend, Tammy Otto. Words will never fully describe the sense of gratitude that I feel to my Babes. v    TABLE OF CONTENTS LIST OF TABLES ............................................................................................................. ix LIST OF FIGURES .......................................................................................................... xii INTRODUCTION .............................................................................................................. 1 CHAPTER 1 USING MULTIPLE METHODS TO ASSESS DETECTION PROBABILITIES OF FOREST-FLOOR WILDLIFE ..................................................................................... 6 Abstract ................................................................................................................... 7 1.1. Introduction ...................................................................................................... 8 1.2. Methods............................................................................................................ 9 1.2.1. Study Area ........................................................................................ 9 1.2.2. Site Selection .................................................................................. 10 1.2.3. Model Generation ........................................................................... 12 1.2.4. Exploring Future Sampling Designs ............................................... 15 1.3. Results ............................................................................................................ 16 1.4. Discussion ...................................................................................................... 19 1.4.1. Management Implications ............................................................... 22 1.5. Acknowledgments.......................................................................................... 23 Literature Cited ..................................................................................................... 36 CHAPTER 2 COMPARING COVER OBJECT AND LEAF LITTER SURVEYS FOR DETECTING RED-BACKED SALAMANDERS, PLETHODON CINEREUS ............ 41 Abstract ................................................................................................................. 41 2.1. Introduction .................................................................................................... 42 2.2. Methods.......................................................................................................... 44 2.2.1. Site Selection .................................................................................. 44 2.2.2. Salamander Searches ...................................................................... 44 2.2.3. Statistical Analyses ......................................................................... 45 2.2.4. Power Approximation ..................................................................... 47 2.3. Results ............................................................................................................ 48 2.4. Discussion ...................................................................................................... 50 2.5. Acknowledgments.......................................................................................... 52 2.6. Literature Cited .............................................................................................. 60 CHAPTER 3 IMPROVING OCCUPANCY ESTIMATION WHEN SAMPLING DISTURBANCES AND ANIMAL MOVEMENTS VIOLATE THE CLOSURE ASSUMPTION ................................................................................................................. 64 Abstract ................................................................................................................. 64 vi    3.1. Introduction .................................................................................................... 65 3.2. Methods.......................................................................................................... 68 3.2.1. Study Area ...................................................................................... 69 3.2.2. Site Selection .................................................................................. 69 3.2.3. Model Development and Analyses ................................................. 71 3.2.4. Simulation Study............................................................................. 75 3.3. Results ............................................................................................................ 77 3.3.1. Salamander Field Study .................................................................. 77 3.3.2. Simulation Study............................................................................. 78 3.4. Discussion ...................................................................................................... 79 3.5. Acknowledgments.......................................................................................... 85 Appendix A.1 Standard Errors for Simulation Results (chapter 3) ...................... 98 Literature Cited ..................................................................................................... 98 CHAPTER 4 SONGBIRD RESPONSE TO GREEN-TREE RETENTION PRESCRIPTIONS IN CLEARCUT FORESTS ............................................................................................ 106 Abstract ............................................................................................................... 106 4.1. Introduction .................................................................................................. 107 4.2. Methods........................................................................................................ 109 4.2.1. Study Area .................................................................................... 109 4.2.2. Site Selection ................................................................................ 110 4.2.3. Bird Sampling ............................................................................... 111 4.2.4. Habitat Covariates ......................................................................... 112 4.2.5. Model Generation and Analysis.................................................... 112 4.3. Results .......................................................................................................... 115 4.4. Discussion .................................................................................................... 118 4.4.1. Interior Forest Birds ...................................................................... 119 4.4.2. Competing Hypotheses ................................................................. 121 4.4.3. Songbird Detection and Movement .............................................. 122 4.4.4. Conclusions ................................................................................... 123 4.5. Acknowledgments........................................................................................ 124 Appendix B.1 Estimating Canopy Cover (chapter 4) ......................................... 126 Literature Cited ................................................................................................... 141 CHAPTER 5 OCCURRENCE AND SURVIVAL OF A TERRESTRIAL SALAMANDER FOLLOWING HABITAT DEGRADATION: CAN WE AMELIORATE THE EFFECT OF TIMBER HARVEST? ............................................................................... 147 Abstract ............................................................................................................... 147 5.1. Introduction .................................................................................................. 148 5.2. Methods........................................................................................................ 150 5.2.1. Study Species ................................................................................ 150 5.2.2. Study Area .................................................................................... 151 5.2.3. Salamander Occupancy ................................................................. 151 vii    5.2.4. Salamander Survival ..................................................................... 153 5.2.5. Data Analysis ................................................................................ 157 5.3. Results .......................................................................................................... 158 5.3.1. Salamander Occupancy ................................................................. 158 5.3.2. Salamander Survival ..................................................................... 159 5.4. Discussion .................................................................................................... 160 5.4.1. Salamander Occupancy ................................................................. 160 5.4.2. Salamander Survival ..................................................................... 161 5.4.3. Probability of Encounter ............................................................... 162 5.4.4. Conclusions ................................................................................... 163 5.5. Acknowledgments........................................................................................ 165 AppendixC.1 Estimating Canopy Cover (chapter 5) .......................................... 167 Literature Cited ................................................................................................... 177 CONCLUSIONS............................................................................................................. 183   viii    LIST OF TABLES Table 1.1.  Small mammals, amphibians, and reptiles detected using drift fence arrays, pitfall traps, visual encounter surveys (VES), and coverboards at 21 sites during 36 trap days in northern Michigan, USA, 2009.…………………….......................................................................................24 Table 1.2.  Parameter estimates for occupancy ( ) and detection probabilities ( ̂ ) for red-backed salamanders (RBSA), northern short-tailed shrews (NSTS), and pygmy and masked shrews (Sorex spp.) detected by all methods combined, visual encounter surveys (VES), pitfall traps, drift fences, and a cover boards (not shown ) at 21 sites and 36 trap days in northern Michigan, USA, 2009. We estimated all parameters from a model assuming constant occupancy and detection probabilities (ψ(.), p(.)).   = the proportion of sites where a species was known to occur (i.e., naïve estimate) .......................................................................................................26 Table 1.3.  Ranking of candidate detection models for 3 wildlife species at 21 sites in northern Michigan, USA, 2009. Covariates include Julian date and 24-hr precipitation for each visit. A squared symbol represents the quadratic term for the respective covariate. We generated models using detection data from drift fences for northern short-tailed shrews and pygmy and masked shrews and visual encounter surveys for red-backed salamanders. All other method-species combinations yielded insufficient detection ˆ probabilities ( p ≤ 0.15, see Results) for inclusion in this analysis. We report the complete model set for red-backed salamanders, but we report only models accounting for 0.90 of the total Akaike weight (wi) for the shrew species. ........................................................................................................28 Table 1.4.  Statistical power approximations for detecting a simulated change in northern short-tailed shrew and pygmy and masked shrew occupancy using 3 alternative sampling scenarios. For all scenarios I simulated detection history data for 2 groups of 25 sites that I visited 4 times in a season. I used the parameter estimates from the highest ranking model for drift fences (Table 3) to represent site occupancy probability for the first group of 25 sites ( ) and for species detection probability ( ̂ ). To simulate a decline in species occupancy, I set the occupancy probability ) to 50% of . Each scenario for the second group of sites ( represents a situation where researchers target sampling effort during days within 24 hr of moderate (10 mm), light (5 mm), or no rainfall (0 mm) ...............31 Table 2.1.  Parameter estimates for occupancy ( ) and detection probabilities ( ̂ ) for Red-backed Salamanders using natural cover object (NCO) and leaf litter surveys at 30 sites in northern Michigan, 2009 (95% confidence interval in ix    parentheses). (obs) is the proportion of sites occupied without accounting for imperfect detection. All parameters were estimated from a model assuming constant occupancy and detection probabilities..........................54 Table 2.2.  Ranking of candidate models for Red-backed Salamanders detected using natural cover object (NCO) and leaf litter surveys at 30 sites in northern Michigan, 2009. Detection covariates include the number of searched NCOs (COVER), survey duration (TIME) and survey type (METHOD). The NULL model assumes detection probability was constant across all sub-plots. ‘Multi-method’ includes data from both surveys types. For all models, occupancy was held constant between sites. Weight (wi) and K represent the probability that model i is the best in the candidate set and number of estimated parameters, respectively. Logit-link parameter estimates are provided for the detection portion of each model (β0 = detection intercept, β1 = detection covariate) ........................................................55 Table 2.3.  Parameter estimates used to simulate a 25% change in salamander occupancy probability and power approximations for detecting the simulated change between two groups of sites using natural cover object (NCO) or leaf litter surveys. Occupancy ( ) and detection probability ( ̂ ) estimates were generated from Table 1, while the maximum sample size ( ) was estimated from my field data. Approximated power is the chance that the false null hypothesis (i.e., no difference in the occupancy probability of g1 and g2) was correctly rejected ...............................................57 Table 3.1.  Selection results for models fit to detection of red-backed salamanders sampled using natural cover object (NCO) surveys in northern Michigan, 2009. ΔAICc represents the difference between AICc values for model i and the top-ranking model; w is the Akaike weight; K is the number of parameters; −2 l is twice the negative log-likelihood ............................................86 Table 3.2.  Parameter estimates for red-backed salamanders sampled using natural cover object (NCO) surveys in northern Michigan, 2009.  βD , , and βC are the beta estimates representing logit-linear effects βT of DISTURBANCE, TREND, and COVER on occupancy and extinction nd probabilities, respectively. β _S is the beta estimate for the 2 survey within a primary period. We report estimates and 95% confidence intervals for each of the top three models (w >0.01) .............................................87 Table 4.1.  Detection probability and standard error estimates ( ± 1SE) for 20 songbird species detected in harvested aspen stands in the northern Lower Peninsula of Michigan, 2010-11. is the probability of detecting the species during a 3 minute point-count, given it was present within a 50m radius of point-count site .....................................................................................135 x    Table 4.2.  The proportion of 222 study sites where the target species was known to occur (i.e. positive detection) during 4, 9-minute point-counts. Each pointcount was separated by 10-14 days. “Combined” represents the proportion of sites where the species was detected at least once during any of the 4 point-counts ............................................................................................137 Table 4.3.  Ranking of candidate occupancy models for 20 songbird species detected at 222 sites in young aspen stands from the northern Lower Peninsula of Michigan, USA, 2010-2011. ‘Retention’ is the percent canopy cover within a 50m radius of a avian point-count site, assigned to 1 of 3 classifications: sites where retention levels were below (<3%), at (310%), or exceeded (>10%) MDNR retention guidelines (Bielecki et al., 2006). ‘Age’ is the number of years post-harvest, assigned to 1 of 3 classifications: 1-4, 5-9, or 10-15 years post-harvest. I report the complete model set for Black-and-white Warblers, but report only models accounting for 0.80 of the total Akaike weight (wi) for all other species. Cumulative model weights (w+) for Retention (Reten) and Age are also reported ................................................................................................................138 Table 5.1.  Ranking of candidate occupancy models for red-backed salamanders detected at 134 sites located in aspen stands from the northern Lower Peninsula of Michigan, USA, 2010-2011. ψ and p represent site occupancy and detection probability, respectively. All models include a “Control” covariate, noting whether a site was located in a recently harvested (1-5 years post-harvest) or 40-60 year-old aspen stand. CANOPY represents the green-tree retention covariate, while CWDi and CWDj represent coarse woody debris covariates for each site and transect within a site, respectively .....................................................................................169 Table 5.2.  Ranking of candidate robust design models for 450 red-backed salamanders within field enclosures deployed in aspen stands, Lower Peninsula of Michigan, 2010-2011. All models include a “Control” covariate, denoting whether the salamander was detected in 1-5 or 40-60 year-old sites. “CANOPY” represents the amount of green-tree canopy cover above each salamander enclosure, while “CWD” represents the number of coarse woody debris objects within each enclosure. p and c represent initial capture and recapture probability of marked salamanders, respectively. b is an additive constant for recapture probability .........................170 xi    LIST OF FIGURES Figure 1.1.  Schematic for a randomly selected 60 × 60-m site with 4 sampling methods to detect (species) in northern Michigan, USA, 2009. Background image represents a sampling lattice with 9 potential sampling sites and shaded areas denote retained canopy green-trees. Symbols represent an array of 4 19-L buckets (circles), 2 1.5 × 0.75-m plywood cover boards (squares), a 20-m visual encounter survey (line), or a drift fence with 5-m wings (sideways “Y”). Each drift fence had paired funnel traps (thin lines) and a bucket at each end and one in the center (circle). I spaced all sampling methods >10 m apart and >10 m from the site boundary.…………………….. .............................................................................33 Figure 1.2.  Estimated detection probability as a function of precipitation within a 24hr period prior to sampling for (A) northern short-tailed shrews and (B) pygmy and masked shrews in northern Michigan, USA, 2009. I calculated detection probabilities using model averaged estimates based on the most supported models. ..................................................................................................34 Figure 1.3.  Estimated detection probability as a function of sampling date for (A) pygmy and masked shrews and (B) red-backed salamanders in northern Michigan, USA, 2009. I calculated detection probabilities using model averaged estimates based on the most supported models ......................................35 Figure 2.1.  Estimated detection probability for Red-backed salamanders modeled as a function of survey duration during natural cover object surveys in northern Michigan, USA, 2009. ...........................................................................................58 Figure 2.2.  Estimated detection probability for Red-backed Salamanders modeled as a function of the number of cover objects that were searched through during leaf litter surveys in northern Michigan, USA, 2009. ............................................59 Figure 3.1.  Sampling lattice of 60 × 60-m cells overlaid on an aspen stand and systematically located sampling transects (in one random cell) for estimating red-backed salamander occupancy, extinction, and detection probability in northern Michigan, USA, 2009. Dark gray polygons in the background image represent unharvested canopy trees following timber harvest. Solid, dashed, and dotted lines in the foreground image represent 20-m transects initially surveyed in May, June, and July, respectively. ................89 Figure 3.2.  Red-backed salamander occupancy probability for transects with (a) high or (b) low levels of natural cover objects (NCO≥ 20 or NCO <20, respectively). Black circles represent Group 1 transects initiated in early May and surveyed through early July; open squares are Group 2 transects xii    initiated in mid June and surveyed through late July; gray triangles are Group 3 transects initiated in late July and surveyed through early August. Each primary period was separated by 7-10 days. All parameter estimates are model averaged ................................................................................................90 Figure 3.3.  Site extinction probabilities for red-backed salamander transects with (a) high or (b) low levels of natural cover objects (NCO≥ 20 or NCO <20, respectively). Black circles represent Group 1 transects initiated in early May and surveyed through early July; open squares are Group 2 transects initiated in mid June and surveyed through late July; gray triangles are Group 3 transects initiated in late July and surveyed through early August. Each primary period was separated by 7-10 days. All parameter estimates are model averaged. ...............................................................................................92 Figure 3.4.  Estimates of relative bias in closed, single-season occupancy models when sites are open to local extinction events between primary periods. = initial occupancy probability, S = the number of primary sampling periods (i) where the number of surveys, t = 1, 2..(2×S), γ = probability an unoccupied site at time i, becomes occupied at time i + 1, ε = probability an occupied site at time i, becomes unoccupied at time i + 1, p = probability of detecting the species during a single survey, given its presence at a site. Expected values were analyzed using time-independent ( . , . ; solid line) and time-dependent ( . , ; dashed line) closed occupancy models. Relative bias was calculated as bias = (E( – /  where E( is the estimated occupancy for the closed model and  is the ‘true’ initial occupancy value. ...........................................94 Figure 4.1.  Number of songbird sampling sites, stratified by green-tree retention level and years since harvest. ‘Green-Tree Retention Level’ represents sites where the proportion of canopy cover within a 50m radius around the point-count station was below (<3%), at (3-10%), or exceeded (>10%) retention guidelines implemented by the MDNR in harvested aspen stands (Bielecki et al., 2006). ‘Years Since Harvest’ is the age of the site at the time sampling occurred ........................................................................................128 Figure 4.2.  Site occupancy estimates for 20 songbird species detected at 222 sites in young aspen stands from the northern Lower Peninsula of Michigan, USA, 2010-2011. Occupancy represents the probability that a site is occupied during a 9min point count. Green Tree Retention (%) is the percent canopy cover within a 50m radius of an avian point-count site; <3, 3-10, and >10% represent green tree retention levels that were below, met, or exceeded MDNR retention guidelines (Bielecki et al., 2006). Years Since Harvest is the age of the forest stand at the time of sampling. All e parameter estimates are model averaged. A = Interior Forest , B-C = f g Generalist , and D = Early-successional species.. ..............................................129 xiii    Figure 4.3.  Site settlement estimates for 20 songbird species detected at 222 sites in young aspen stands from the northern Lower Peninsula of Michigan, USA, 2010-2011. Settlement is the probability that an unoccupied site during primary period i becomes occupied before primary period i + 1. Green Tree Retention (%) is the percent canopy cover within a 50m radius of an avian point-count site; <3, 3-10, and >10% represent green tree retention levels that were below, met, or exceeded MDNR retention guidelines (Bielecki et al., 2006). Years Since Harvest is the age of the forest stand at the time of sampling. All parameter estimates are model averaged. A e f g = Interior Forest , B-C = Generalist , and D = Early-successional species. ......132 Figure 5.1.  Occupancy probability for red-backed salamanders in 1-5 year-old clearcut aspen stands in the northern Lower Peninsula of Michigan, 20102011, explained as a function of (A) percent of site area that was covered by green tree canopy retention, and (B) the number of coarse woody debris objects at each site. All estimates were generated using model averaging. .............................................................................................................171 Figure 5.2.  Estimated salamander survival probability in the northern Lower Peninsula of Michigan, 2010-2011, explained as a function of (A) percent residual canopy cover found above a field enclosure, (B) the number of coarse woody debris objects within a field enclosure, (C) or both variables. Survival is the probability that a marked salamander i survived 1 month in a field enclosure located in an aspen stand that was 1-5 years post harvest. All estimates were generated using model averaging.....................173 Figure 5.3.  Model-averaged estimates of initial capture (p) or recapture (c) probability of salamanders during 3 sampling events (visits) in mid-June, northern Lower Peninsula of Michigan, 2010-2011. Initial capture (pt) is the probability that a marked salamander is captured for the first time during visit t, conditional on it surviving and being available for capture. Recapture (ct) is the probability a marked salamander is recaptured during visit t, conditional on it being captured at least once before during a previous visit ........................................................................................................176 xiv    INTRODUCTION Ecologists employ a systematic and objective system of inquiry to infer relationships between organisms and their environment. Although they strive to achieve “strong inference” in their research (Platt 1964), judicious ecologists recognize that research results are seldom clean: stochasticity and process variability are phenomenon that must be acknowledged, if not openly embraced. This requires ecologists, willingly or otherwise, to venture into the worlds of quantitative modeling, probability theory, statistical inference, and hypothesis development. Ecologists must be strategic in the development of their a priori research hypotheses, represent those hypotheses by statistical models, and carefully design a study, or studies, that will allow them to assess the evidence for these hypotheses, given the observed data. The overarching theme of my dissertation is one of conservation-based research for understanding the impacts of anthropogenic disturbances, specifically timber harvesting, on forest wildlife. To the best of my ability, I have adopted scientific approaches that have been advocated by Chamberlin (1965), Platt (1964), and Anderson (2008), among others. All five of my dissertation chapters involve observational wildlife data, and hence possesses some degree of observation sampling error. I view the data for my dissertation as generated from two distinct processes: biological and observational. The biological process is of interest to most ecologists and conservation biologists. Examples include the effect of habitat alteration on a particular species or annual changes in population growth rate. However, biological reality is often obfuscated by observational sampling error, such that the true state of the study system is difficult to observe. For most ecologists, sampling error represents statistical “noise”, which needs to be accounted for, so that biological reality may be viewed more clearly. 1    Detectability is a fundamental, and yet often overlooked, issue when modeling species distribution patterns. For example, a species may be present at a study site and yet go undetected during routine sampling events. For most animals, natural selection has favored traits that allow the organisms to conceal themselves within their environment. Thus, it is of no surprise to most field-savvy naturalists that many wildlife species are difficult to detect. However, many ecologists have ignored the issue of detection and thus wide-spread confusion exists regarding what variable is actually being estimated in species distribution models (Yoccoz et al. 2001). This point can be summarized succinctly via the equation E(C) = N × p, where E(C) is the expected count of sites where species x was observed (i.e., the field data), N is the true number of occupied sites (which is unknown to the researcher), and p is the probability of detecting species x, given it occurs at the site. After rearranging the above equation to clear that obtaining unbiased estimates of = E(C)/ ̂ it becomes , require inferences about ̂ . Thus, distribution models that do not account for imperfect detection actually estimate apparent occurrence or effective detection probability: the combination between true occurrence of species x and the probability of detecting species x, given it is present. A similar line of reasoning can be applied to estimating animal abundance or population vital rates (i.e., survival, fecundity, movement). Species distribution modeling is a common theme of my dissertation; hence, I felt it prudent to use both design- and model-based approaches to account for the limited detectability of my target wildlife species. Indeed, my first three dissertation chapters are dedicated to improving inferential power in wildlife research when the target species are difficult to detect. I used the results from Chapters 1-3 to develop a strong study design for assessing the effectiveness of structural retention for conserving wildlife in Michigan’s harvested forests (Chapters 4 and 5). 2    For my first two chapters I used species occupancy models and multiple sampling techniques to estimate detection probabilities of small mammals, amphibians, and reptiles (i.e., forest-floor wildlife). Specifically, I assessed the influence of survey method, seasonality, and precipitation on detection probability of forest-floor wildlife and explored study design tradeoffs using parameter estimates from my primary analyses and simulated data. These chapters highlight the importance of coupling design- and model-based approaches for reducing false absences, and highlight the utility of pilot data for designing wildlife research projects which yield strong inferences. I used results from these analyses to inform the sampling design for Chapter 5. My third chapter involves a detailed investigation into performance of wildlife occupancy models when model assumptions are knowingly violated. In this study I investigated potential bias associated with two processes that violate the closure assumption (i.e., no colonization or extinction) during a sampling season: 1) repeated destructive sampling events that result in either short- or long-term site avoidance by the target species and 2) sampling occurring over a time period during which non-random movements of the target species result in variable occupancy status. I used dynamic occupancy models to quantify the potential bias in occupancy estimation associated with these processes for a terrestrial salamander system. I generalized my findings by conducting a simulation to evaluate how violations of closure can bias occupancy estimates when organism availability declines within a field season. This chapter highlights the importance of addressing closure and non-random organism movements in wildlife occupancy studies. I provide multiple solutions for minimizing bias associated with non-random changes in occupancy and repeated sampling disturbances. 3    Chapters 4 and 5 represent the impetus for my dissertation research: Assessing the role of structural retention for conserving wildlife in harvested forests. Both studies represented an empirical evaluation of the Michigan Department of Natural Resources (MDNR) structural retention guidelines. In Chapter 4 I evaluated the effectiveness of green-tree retention for promoting site occupancy of interior forest, generalist, and early-successional songbirds in aspen stands that were 1 to 15 years post-harvest. I provide multiple management recommendations for the future use of structural retention in actively managed landscapes. In Chapter 5 I used correlative and experimental evidence to evaluate how retention of vertical green trees (CANOPY) and coarse woody debris (CWD) influenced terrestrial salamander occupancy and survival in clearcut forests. First, I studied how patterns in salamander occupancy were related to retention of green trees and coarse woody debris within a managed, forested landscape (560,000ha). Second, I quantified how salamander survival was influenced by CANOPY and CWD within timber harvest areas. By focusing my study on the same species, forested areas, and time period I was able to evaluate if occupancy and survival measurements yielded similar inferences regarding amphibian response to structural retention. Thus, this study informs the current debate over which population parameters are most useful for assessing amphibian response to habitat degradation and highlights the importance of coupling broad-scaled distribution research with detailed demographic studies. 4    LITERATURE CITED 5    LITERATURE CITED Anderson, D.R. 2008. Model Based Inference in the Life Sciences: A Primer on Evidence. Springer, New York, New York, USA. Chamberlain, T.C., 1890. The method of multiple working hypotheses. (Reprinted 1965, Science, 148:754-759). Platt, J.R., 1964. Strong inference. Science 146, 347-353. Yoccoz, N.G., Nichols, J.D., Boulinier, T., 2001. Monitoring of biological diversity in space and time. Trends in Ecology and Evolution 16, 446-453. 6    CHAPTER 1 USING MULTIPLE METHODS TO ASSESS DETECTION PROBABILITIES OF FOREST-FLOOR WILDLIFE Abstract Many previous comparisons of multiple sampling methods have assumed that detection probabilities for each method are either constant or equal to one. I used 4 sampling methods to estimate detection probabilities for forest-floor dwelling amphibians, reptiles, and small mammals. I investigated associations between seasonality and precipitation on species detection and explored sample design tradeoffs for future studies. Although I captured 25 species, I could reliably detect (detection probability >0.15) only northern short-tailed shrews (Blarina brevicauda) and pygmy and masked shrews (Sorex spp.) using drift fences and red-backed salamanders (Plethodon cinereus) using visual encounter surveys (VES). The use of multiple sampling methods improved detection probabilities for only red-backed salamanders ( ̂ 0.32, 95% CI: 0.24−0.38,  ̂   = = 0.38, 95% CI: 0.32−0.44). Parameter estimates indicated detection of both shrew species was positively related to increased precipitation. Detection probabilities for pygmy and masked shrews and red-backed salamanders were positively and negatively associated with date, respectively. My power analysis revealed that sampling during rain events increased the power of detecting a change in sorid occupancy by ≥40% (α = 0.05). My results demonstrate the need to incorporate species detectability when comparing the effectiveness of different trapping methodologies. Furthermore, my study highlights the utility of power analyses for exploring study design tradeoffs for research and monitoring programs. 7    Keywords: amphibians, detection probability, Michigan, multiple methods, occupancy, precipitation, reptiles, small mammals. Status: Published. Otto, C.R.V., Roloff, G.J., 2011. Using Multiple Methods to Assess Detection Probabilities of Forest-Floor Wildlife. Journal of Wildlife Management 75, 423-431 1.1. Introduction Failure to account for imperfect detection has been strongly criticized as a shortcoming of endangered species conservation, forest management, and wildlife monitoring (Yoccoz et al. 2001, MacKenzie et al. 2005, Kéry and Schmidt 2008, Kroll 2009). Indeed, detection of most forest-floor vertebrates, such as terrestrial amphibians, reptiles, and small mammals, is likely confounded by seasonality, weather, habitat, survey type, and effort. Detection probabilities of these taxa may be further confounded if only a small fraction of the population is available for sampling (i.e., fossorial behavior patterns; Bailey et al. 2004a). Hence, imperfect detection can lead to biased or erroneous conclusions about population and community parameters (Mazerolle et al. 2005, Jackson et al. 2006). Several authors suggest using multiple survey methods to reduce or eliminate statistical noise caused by poor detection probability (e.g., Ryan et al. 2002, Garden et al. 2007, RibeiroJúnior et al. 2008, Hutchens and DePerno 2009). However, the assumption that multiple methods alleviate biases associated with poor detection is rarely tested. One negative aspect of employing multiple methods is the potential to collect redundant information if one method performs better than all others, adding to costs for time and materials (Mattfeldt and Grant 2007). Also, the use of multiple methods could lead to erroneous conclusions if all methods perform poorly. Past multi-method studies that accounted for imperfect detection have been limited to salamanders and meso- and large mammals (Bailey et al. 2004b, O’Connell et al. 2006, Mattfeldt 8    and Grant 2007). I utilized detection data collected on plethodontid salamanders, reptiles, and small mammals (hereafter forest-floor wildlife) in northern Michigan, USA to estimate species detection probability and occupancy using multiple survey methods. My objectives were to 1) assess the influence of survey method, seasonality, and precipitation on detection probability of forest-floor wildlife and 2) explore future study design tradeoffs using parameter estimates from both my analysis and simulated population data. Specifically, I compared the ability of 3 competing study designs to detect a 50% change in the proportion of sites occupied by northern short-tailed shrews (Blarina brevicauda) and pygmy and masked shrews (Sorex spp.). 1.2. Methods 1.2.1. Study Area I conducted my study in the Cadillac-Traverse City area in the northwestern Lower Peninsula of Michigan, USA, in 2009. This area occurs in a glacial outwash-plain with porous sandy soils, intermixed with ice-contact zones and moraines (Barnes and Wagner 2004). My study took place on state-owned lands with a history of timber harvest that were managed for aspen (Populus sp.) production at the time of my study. I sampled 3 forest stands that were 5-8 yr postharvest and 15-28 ha in size. Within each stand the Michigan Department of Natural Resources and Environment (MDNR) implemented variable-retention harvest prescriptions (Bielecki et al. 2006), which call for retention of 3-10% of live tree basal area and retention of vertical snags and coarse woody debris (CWD) whenever possible. Retained canopy trees consisted of mixed oak (Quercus spp.), eastern white pine (Pinus strobus), red maple (Acer rubrum), black cherry (Prunus serotina), American beech (Fagus grandifolia), and aspen, whereas dominate understory woody plants were aspen, red maple, blackberry (Rubus spp.), black cherry, downy serviceberry (Amelanchier arborea), witch-hazel (Hamamelis virginiana), 9    American beech, and hophornbeam (Ostrya carpinifolia). Prior to European settlement, this ecosystem was dominated by eastern white and jack pine (Pinus banksiana) and prone to fire disturbances (Barnes and Wagner 2004). 1.2.2. Site Selection I used HAWTH’S TOOLS (Hawth’s Tools, version 3.27, http://www.spatialecology.com/htools/, accessed 1 Feb 2009) in a Geographic Information System (ArcGIS 9.1) to overlay each stand with a sampling lattice, consisting of 60 × 60-m cells, to delineate potential study sites. I recognized that the characteristics of the surrounding landscape could introduce heterogeneity in wildlife occupancy (Prugh et al. 2008). Therefore, I eliminated all lattice cells whose borders intersected or encompassed an unharvested forest edge, active logging road or off-road recreational vehicle trail, or wetland appearing on 2005 National Agricultural Imagery Program imagery (Michigan Department of Information Technology 2009). I considered all remaining lattice cells as potential sampling locations and randomly selected 7 within each forest stand. My selection process produced 21 sampling sites, spaced ≥60 m apart. I chose 7 sites per stand to ensure I had adequate replication of sites at each stand while still allowing for sufficient site separation. Within each 60 × 60-m site, I used 4 techniques to detect forest-floor wildlife: drift fence arrays, pitfall arrays, coverboard arrays, and visual encounter surveys (VES). I systematically arranged the various detection methods with a drift fence at the center, a coverboard or pitfall array placed equidistant from the drift fence array in the 4 corners of the site, and paired VES transects at the north and south ends (Figure 1.1). I positioned Y-shaped drift fences along an east-west axis (Figure 1.1). Each segment of the fence was 5 m long. I constructed fences of silt fencing (45 cm tall) and buried them 15 cm in the ground with one 19-L bucket buried at each 10    end (Figure 1.1). I buried another bucket where the 3 wings converged. I placed a pair of 1-m double-ended funnel traps flush with the ground along each drift fence wing (Figure 1.1). Funnel traps had 20-cm and 5-cm outer and inner openings, respectively (Corn 1994).   For each pitfall array, I buried 4 19-L buckets in the shape of a diamond and spaced them approximately 1 m apart (Figure 1.1). I drilled holes in the bottoms of the buckets to facilitate water drainage and placed a piece of carpet in each bucket to provide shading. A coverboard array (Fellers and Drost 1994, Marsh and Goicochea 2003) consisted of 2 1.5 × 0.75-m plywood boards laid flush with the ground, spaced 1 m apart (Figure 1.1). The VES (Crump and Scott 1994) consisted of a 20-m transect set at 20 m from the north and south ends of the drift fence array (Figure 1.1). For each VES, one observer searched for wildlife under CWD >4 cm diameter, >15 cm long, and within 1 m of the transect centerline. Wildlife observers were rotated among transects to minimize observer bias in detection (i.e., no observer surveyed the same transect twice in a row). I positioned pitfall and coverboard arrays and VES 10 m from the edge of a site’s boundary (Figure 1.1). I sampled sites from 5 May 2009–2 July 2009. I adopted a rotating sampling design where I sampled all sites within one forest stand twice in a 24-hr period (i.e., 2 visits). I opened traps 24 hr prior to the first visit and closed them after the second visit. I then commenced sampling at another forest stand and visited all sites at this stand twice. I rotated sampling effort between 3 stands until all sites had 12 visits. I ensured that primary sampling events were 7-10 days apart (i.e., visits 1-2, 3-4, 5-6, ect. were each 7-10 days apart). During each visit, I collected species detection information for all methods and released organisms >5 m from the point of capture for drift fence arrays and pitfalls. I identified all organisms to species with the exception of pygmy (S. hoyi) and masked shrews (S. cinereus), which I could only identify to 11    genus. My sampling and handling protocols were approved by the Michigan State University Animal Care and Use Committee (Animal Use Form no. 07/08-118-00). At each forest stand, I used 4 rain gauges to determine average precipitation within a 24hr period prior to each sampling visit. I also recorded the Julian date of each visit. For VES transects, I quantified the number of cover objects rolled during each visit and the survey duration. 1.2.3. Model Generation I used single-season models to estimate species occupancy and detection probabilities at a site (MacKenzie et al. 2002). The single-season model allowed for estimation of occupancy for species whose detection probability was <1. For all analyses, I held species occupancy probability constant (ψ(.)), which assumes that habitat heterogeneity does not have an appreciable effect on species occurrence at a site. This assumption was robust for the 3 species I detected effectively (i.e., only northern short-tailed shrews, pygmy and masked shrews, and redbacked salamanders had ̂ ≥ 0.15; see Results). Site occupancy probabilities were ≥0.75 for all 3 species, suggesting they were relatively ubiquitous across my study sites. Therefore, the influence of habitat covariates on the probability of site occupancy would likely be minor. I pooled detection data for all methods and constructing a model for each species that assumed constant occupancy and detection probabilities. This model estimated the probability of detecting species x with ≥1 survey method during one visit, given the species was present at the site. I used this analysis to determine which species I would remove from subsequent analyses, as reliable estimates of occupancy cannot be calculated for species with very low detection probabilities (i.e.,  ̂ ≤ 0.15). I predicted detection probability of pond-breeding amphibians would be ≤ 0.15 because my study sites were not near wetlands. However, I included pond- 12    breeding amphibians in my initial analysis because other multi-method studies have captured pond-breeding amphibians in terrestrial habitat (Ryan et al. 2002, Roloff et al. 2010). Furthermore, use of terrestrial habitat by pond-breeding amphibians has been increasingly documented (Semlitsch and Bodie 2003, Rittenhouse and Semlitsch 2007, Rittenhouse et al. 2009). I felt it was reasonable for most small mammal, squamate, and plethodontid species to have  > 0.15 for this initial analysis. For species with ̂  > 0.15, I analyzed data from each survey method independently, to evaluate how different survey methods may influence bias and precision of detection and occupancy estimates. I predicted that coverboard arrays would perform well for sampling plethodontids and squamates (Monti et al. 2000, Marsh and Goicochea 2003, Bailey et al. 2004b). I recognized that microclimate conditions developing underneath coverboards would dictate species utilization: whereas desiccation-prone species are less likely to occur under a coverboard if the habitat remains dry, this circumstance would be ideal for squamates. I predicted VES would provide high detection probabilities for sedentary species such as plethodontids but not for more mobile organisms such as squamates and small mammals (Smith and Petranka 2000). I expected drift-fences would perform well in detecting more mobile organisms, as well as plethodontids during rain events (Ryan et al. 2002). I considered pitfall traps as a cheaper (both in installation time and money) alternative to drift fences and predicted they would be effective for detecting small mammals, terrestrial salamanders, and small squamates (Butts and McComb 2002, McDade and Maguire 2005, Garden et al. 2007). I eliminated method-specific models with ̂ ≤ 0.15 for individual species from subsequent analyses (see O’Connell et al. 2006). 13    For all remaining species, I constructed method-specific models that allowed detection probability to vary as a function of Julian date and precipitation within a 24-hr period. I explored potential correlation between Julian date and 24-hr precipitation using Pearson’s product moment correlation (Sokal and Rohlf 1981). I predicted that Julian date would positively associate with detection probability of squamates, negatively associate with plethodontid detectability, and exhibit no association with shrew detectability (Gibbons and Semlitsch 1987, Kurta 1995, Petranka 1998:342, Ford et al. 2002). I predicted that rain events would increase the probability of detecting plethodontid salamanders and shrews due to their increased surface activity during wet weather (Petranka 1998:342, Umetsu et al. 2006). I constructed linear and quadratic terms for each covariate, and constructed additive models that included precipitation and Julian date together, yielding 8 potential a priori models for each species and method. I also conducted an exploratory analysis to see if red-backed salamander detection was negatively associated with observers repeatedly performing VESs. During my study, I noticed that repeated handling of CWD along VES transects resulted in fragmentation of those objects during later visits (ti). Therefore, I used the change in cover object number as a surrogate for observer-induced disturbance and modeled its influence on salamander detection probability. I used the number of cover objects tallied during the first visit as my baseline (t0) and calculated the difference in cover objects at ti and t0 as my measure of observer-induced disturbance. This calculation provided an estimate of cover object fragmentation during each visit and for each site. I developed one model with this variable as a detection covariate and compared it to my a priori model set. 14    I generated parameter estimates via maximum likelihood estimation using Program PRESENCE (PRESENCE, version 2.3, http://www.mbr-pwrc.usgs.gov/software/presence.html, accessed 1 May 2009). I used Akaike’s Information Criterion, adjusted for small sample size (AICc), and model weights to rank models (Burnham and Anderson 2002). I used model averaging and unconditional variance estimation when multiple models received support in the analysis (i.e., ΔAICc ≤ 10.0, Burnham and Anderson 2002). To account for potential overdispersion in model selection, I used Pearson’s chi-square as a goodness-of-fit test for the global model of each species (MacKenzie and Bailey 2004). I used a parametric bootstrap procedure with 10,000 iterations to estimate the overdispersion parameter ( ̂ ). I used Quasi Akaike’s Information Criterion (QAICc; Burnham and Anderson 2002) for model selection and inflated the standard error of the parameter estimates by √ ̂ if there was evidence that the global model fit the data poorly (i.e.,  ̂ >1; MacKenzie et al. 2006). I reported 95% confidence intervals for all parameter estimates, unless otherwise stated. 1.2.4. Exploring Future Sampling Designs I used parameter estimates generated from my top ranking drift fence model to explore how sampling effort can be allocated in future studies to maximize power for detecting a 50% decrease in the proportion of sites occupied by northern short-tailed shrews and pygmy and masked shrews. I developed 3 potential study designs that included 2 groups of 25 sites with 4 replicate visits. For the first design, I used detection estimates for each species during non-rain events (i.e., 0 mm precipitation within the past 24 hr). The first design represented a situation where researchers sample for sorids without targeting rain events. For the final 2 designs, I used detection estimates from sampling visits with 5 mm and 10 mm of precipitation, respectively. 15    The final 2 designs represented situations where researchers are able to focus their sampling effort during periods of rain. I used the estimate of occupancy from the top ranking model as my baseline for the first group of 25 sites and multiplied this value by 50% for my estimate of occupancy for the second group. I used Program GENPRES (GENPRES, http://www.mbrpwrc.usgs.gov/software/presence.html, accessed 1 May 2009) to simulate detection history data using the parameter estimates described above. I constructed competing models in Program MARK (MARK, version 5.1, http://www.cnr.colostate.edu/~gwhite/mark/mark.htm, accessed 1 Aug 2009) using the simulated data. I compared a null model (H0 = ψ(.), p(.)) that assumed occupancy was constant between the 2 groups to an alternative model (Ha = ψ(group), p(.)) that allowed occupancy to vary between groups. In all cases, I considered Ha the true model, as I used this model to generate the data. I compared the null and alternative models using a likelihood ratio test (LRT) in MARK and approximated power using the chi-square statistic from the LRT output (Burnham et al. 1987, Bailey et al. 2007). This analysis allowed us to approximate the power of a proposed sampling design for detecting a simulated change in species occupancy. 1.3. Results I detected 659 organisms representing 9, 9, and 7 species of small mammals, amphibians, and reptiles, respectively (Table 1). Drift fence arrays detected the most species (22), followed by pitfall traps (11), VES (10), and coverboards (7; Table 1). Estimated detection probabilities varied between 0.01−0.38 when I combined capture data for all methods (Table 2). Red-backed salamander, northern short-tailed shrew, and pygmy and masked shrews were the only species 16    with ̂ ≥ 0.15. The method-specific analysis revealed that pitfall traps and coverboards alone could not reliably detect any species ( ̂  < 0.15), whereas VES were sufficient for detecting only red-backed salamanders ( ̂  = 0.32, 95% CI: 0.26 - 0.38; Table 2). Detection probabilities for drift fence arrays were <0.15 for all species except pygmy and masked shrews ( ̂ = 0.24, 95% CI: 0.18 - 0.30). However, I also included northern short-tailed shrews for further analysis because of a borderline detection probability ( ̂ = 0.14, 95% CI: 0.08 - 0.20) when using drift fences. Julian date and 24-hr precipitation were not correlated (n = 36, r = 0.05, 95% CI: −0.28 0.37), which allowed us to treat them as independent detection covariates. Models containing the 24-hr precipitation covariate received substantial support for both shrew species detected with drift fences (Table 3). As predicted, precipitation was positively related to detection of both sorid species (Figure 1.2). Models that included a quadratic precipitation term received >95% and 70% of the total model weight for northern short-tailed shrews and pygmy shrews, respectively. Detection probabilities for both of these species were lower during dry periods and heavy rain events (Figure 1.2). Detection probability increased as the season progressed for pygmy and masked shrews ( = 1.04, 95% CI: 0.54 - 1.54; Figure 1.3a). Furthermore, some support existed for models containing the quadratic form of Julian date, which suggested pygmy and masked shrew detection decreased during the later part of my study ( 2 = −0.02, 95% CI: −0.01 - −0.03; Figure 1.3a). I found less support for models containing Julian date on northern short-tailed shrew detection (ΔAICc = 3.1; Table 3). 17    2 I found evidence of poor model fit for red-backed salamanders (χ = 223.3, P = 0.02). Therefore, I used QAICc for model selection and inflated the standard error of the parameter estimates by √ ̂ = 1.88. Top model weights indicated that Julian date was associated with redbacked salamander detection (Table 3) with detection probability decreasing as the season progressed (Figure 1.3b). However, there was also support for the constant detection model (ΔQAICc = 1.97, wi = 0.20) and model-averaging results revealed the 95% confidence interval for the Julian date beta estimate approached zero ( = −0.63, 95% CI: −0.02 - −1.24). Counter to my prediction, models containing precipitation showed little support for being associated with salamander detection. My exploratory analysis revealed that the model containing the change in the number of cover objects covariate, my surrogate for habitat disturbance, ranked higher than all other models in my candidate set and received 57% of the total model weight. As predicted, salamander detection probability was negatively associated with the number of fragmented cover objects along a transect ( = −0.49, 95% CI: −0.15 - −0.83). The next highest ranking model included Julian date (ΔQAICc = 1.77 wi = 0.23). My simulation revealed variation in the ability of the 3 proposed study designs to detect a 50% change in occupancy of northern short-tailed shrews and pygmy and masked shrews (Table 4). For both species, designs that focused sampling effort during rain events yielded the highest approximated power. Sampling during non-rain events provided moderate power for detecting a change in pygmy and masked shrew occupancy (40%, α = 0.05) but virtually no power for detecting a change in northern short-tailed shrew occupancy (5%, α = 0.05). Although estimated detection probabilities for northern short-tailed shrews varied during events with 5 mm or 10 mm 18    of precipitation (Figure 1.2a), it did not appear to influence approximated power of the 2 study designs (61% and 63%, respectively, α = 0.05). 1.4. Discussion My study corroborates a growing body of literature highlighting the importance of accounting for imperfect detection in wildlife studies (e.g., O’Connell et al. 2006, Mazerolle et al. 2007, Kéry and Schmidt 2008, Kroll 2009). Some researchers advocate use of multiple sampling techniques to help ameliorate the effects of variable detection (e.g., Garden et al. 2007, RibeiroJúnior et al. 2008, Hutchens and DePerno 2009). However, using multiple methods alone does not address the problem of imperfect detection when all methods perform poorly or if one method influences another when multiple methods are used simultaneously (Nichols et al. 2008). With the exception of red-backed salamanders, multiple detection methods did not improve detection probabilities for any species. In most wildlife studies, implementation of multiple methods may be constrained by limited sampling effort and higher equipment costs (Garden et al. 2007, Ribeiro-Júnior et al. 2008). My results suggest that implementing one detection method based on the behavior and ecology of the target species and accounting for imperfect detection through estimation may be more cost-effective than implementing multiple methods. I was unable to reliably detect any of the 7 species I observed using coverboards, which was counter to my prediction, given coverboards are extensively utilized to detect herpetofauna species, especially terrestrial salamanders and squamates (e.g., Monti et al. 2000, Ryan et al. 2002, Marsh and Goicochea 2003, Bailey et al. 2004b). Monti et al. (2000) suggested that coverboards should age for ≥1 yr to allow the ideal microclimate to develop underneath and also allow time for organisms to encounter the artificial refugia. Indeed, the microclimate under many of my coverboards remained dry, which likely hindered their ability to provide refuge for 19    salamanders but not for squamates. Strangely, the aging period of my coverboards was similar to that of other studies that successfully used this method to capture these organisms (e.g., Hyde and Simons 2001, Ryan et al. 2002, Marsh and Goicochea 2003). Bailey et al. (2004b) found coverboards provided >0.15 detection probabilities for 6 out of 7 terrestrial salamander species. However, coverboards used by Bailey et al. (2004b) were aged 2-4 yr prior to sampling (L. L. Bailey, Colorado State University, personal communication), which may explain discrepancies between detection estimates in Bailey et al. (2004b) and my study. Similar to my results for coverboards, I could not reliably detect any of the 11 species I captured using pitfall traps. Pitfall traps have been used to sample a variety of forest-floor wildlife (Butts and McComb 2002, McDade and Maguire 2005, Garden et al. 2007). To my knowledge, no other study has questioned the assumption of perfect detection probability at the species level when using pitfall traps. Of the 22 species I captured using drift fences, only northern short-tailed shrews and pygmy and masked shrews had detection probabilities ≥0.15. Detection probabilities for those species were positively associated with precipitation events, a phenomenon documented for some shrew species (Brannon 2002, Umetsu et al. 2006) but not others (Laakkonen et al. 2003, Whittaker and Feldhamer 2005). I was 3 times more likely to detect pygmy and masked shrews and 30 times more likely to detect northern short-tailed shrews when sampling on days with 10 mm of precipitation within a 24-hr period compared to days without rain (Figure 1.2). Counter to my prediction, the top pygmy and masked shrew model also included Julian date as a covariate, with detection probability increasing as the season progressed. Although pygmy and masked shrews are known to be active year-round, they may also increase foraging activity as more insect prey become available during the early summer (Kurta 1995). 20    Although I detected 10 species using VES, only red-backed salamanders had a constant detection probability >0.15. I found evidence that my global model for red-backed salamanders fit the data poorly; therefore I exercised caution when interpreting covariate effects. I found some support for a negative association between seasonality and salamander detection probability, which could result from reduced surface activity during summer (Petranka 1998 and references therein). Alternatively, the observed decrease in detection probability could result from disturbances associated with repeated transect visitation. My exploratory analysis revealed a decrease in salamander detection probability at sites where the number of cover objects increased over time, which may be attributed to an increased fragmentation of CWD caused by repeated observer handling. The notion that invasive sampling techniques, such as VES, can influence population parameters through repeated habitat disturbances has been proposed but not empirically tested (Bailey et al. 2004b). More research is needed to determine how salamander occupancy and detection probabilities may be impacted by seasonality and observer-induced disturbances. Based on current information, I suggest sampling effort be allocated in May-June for environments similar to what I sampled in northern Michigan and the number of replicate surveys should be ≤4 to minimize site disturbance. I acknowledge a potential lack of method independence resulting from animals captured in drift fences and pitfall traps not being available for detection by my other methods during a daily sampling event. Violation of the independence assumption would result in underestimated detection probability for any method where a species was unavailable. Although I did not explicitly test for trap independence, I think the probability of violating this assumption was acceptably low for my study species because they exhibit high local abundances and low daily movement rates relative to the distance between my sampling methods (Heatwole 1962, 21    Kleeberger and Werner 1982, Kurta 1995, Getz and McGuire 2008). Furthermore, I reliably detected only 3 species when I combined detection data for all methods, suggesting that potential method dependence did not influence my overall conclusions. Nichols et al. (2008) recommended a multi-method occupancy model, which accounts for a lack of independence between methods. However, low detection estimates for most methods in my study precluded its use. To my knowledge, only O’Connell et al. (2006) and Nichols et al. (2008) have acknowledged method dependence in multi-method studies. However, I am not aware of any research that has quantified this potential source of bias. One benefit of using occupancy modeling is the availability of simulation methods for assessing the statistical power of proposed study designs (Bailey et al. 2007, Mattfeldt et al. 2009). My simulation study revealed that not all sampling designs were equal in their ability to detect a change in sorid occupancy. Although my simulation study was based on overly simplistic sampling designs, it does highlight how researchers can use pilot data to improve sampling efficiency. In this case, rather than using pitfall traps and drift fences for detecting sorid species during random days throughout a season, a more efficient design would include a focused sampling effort during rain events using only drift fences. When pilot data are not available, researchers can still explore design trade-offs using parameter estimates from other studies or best professional judgment (MacKenzie and Royle 2005, Bailey et al. 2007). 1.4.1. Management Implications My results do not support the assumption from past multi-method studies that using multiple sampling techniques will ameliorate the effects of poor detection. As a result, I recommend that studies and monitoring efforts use pilot data to estimate detection probabilities of targeted species using multiple methods and use this information as a basis for selecting a primary 22    sampling method for future studies. As part of this process, I recommend that managers and researchers explore sampling design tradeoffs to ensure the proposed design has sufficient power for detecting real population trends. Pilot data can be used to investigate study designs tradeoffs for detecting changes in species occupancy and improve the allocation of sampling resources for monitoring and managing populations of concern. 1.5. Acknowledgments I thank Andrew Coleman and Rachelle Sterling for their assistance in the collection of field data and Tammy Otto and Jordan Roloff for aiding with sampling installation. Mike Donovan and Kerry Fitzpatrick (MDNR) helped develop project objectives and Allan O’Connell, A. J. Kroll, and 2 anonymous reviewers provided insightful comments. Support for this project was provided by the MDNR Wildlife Division with funds from the federal Pittman-Robertson Wildlife Restoration Act grant administered by the United States Fish and Wildlife Service (W-147-R: Michigan’s Statewide Wildlife Research and Restoration Program) and the Michigan Society of Herpetologists. 23    Table 1.1. Small mammals, amphibians, and reptiles detected using drift fence arrays, pitfall traps, visual encounter surveys (VES), and coverboards at 21 sites during 36 trap days in northern Michigan, USA, 2009. Species Scientific name Drift fence Pitfall VES Coverboard Mammals Eastern chipmunk Tamias striatus x Meadow vole Microtus x Blarina brevicauda x x x Peromyscus sp. x x x Pygmy and masked shrew Southern bog lemming Sorex spp. x x Synaptomys cooperi x x Southern red-backed vole Clethrionomys gapperi x x Star-nosed mole Condylura cristata x Woodland vole Microtus pinetorum x American toad Anaxyrus americanus x Eastern gray treefrog Hyla versicolor Green frog Lithobates clamitans x Northern leopard frog Lithobates pipiens x Plethodon cinereus x Notophthalmus x x Ambystoma maculatum x x a Northern short-tailed shrew Peromyscus sp. a x x x x x Amphibians Red-backed salamander Red-spotted newt Spotted salamander a x 24    x Table 1.1 (cont’d) Species Scientific name Drift fence Pitfall VES Coverboard Spring peeper Pseudacris crucifer x x x Wood frog Lithobates sylvaticus x x x Brown snake Storeria dekayi x x Eastern garter snake Thamnophis sirtalis x x Five-lined skink Plestiodon faciatus x Hog-nosed snake Heterodon platirhinos Red-bellied snake Storeria x Ring-necked snake Diadophis punctatus x Smooth green snake Liochlorophis vernalis Reptiles No. of species a x x x x 22 11 10 7 Species that surpassed the minimum detection threshold of 0.15 when modeled using site-occupancy estimation and which we considered for further analysis. 25    x x Table 1.2. Parameter estimates for occupancy ( ) and detection probabilities ( ̂ ) for red-backed salamanders (RBSA), northern shorttailed shrews (NSTS), and pygmy and masked shrews (Sorex spp.) detected by all methods combined, visual encounter surveys a (VES), pitfall traps, drift fences, and cover boards (not shown ) at 21 sites and 36 trap days in northern Michigan, USA, 2009. We estimated all parameters from a model assuming constant occupancy and detection probabilities (ψ(.), p(.)).   = the proportion of sites where a species was known to occur (i.e., naïve estimate). All methods combined VES transect Species ̂ SE ̂ RBSA 0.38 0.03 1.00 1.00 0.00 NSTS 0.15 0.03 0.76 0.91 Sorex spp. 0.29 0.03 1.00 1.00 ̂ SE ̂ 0.32 0.03 0.13 0.00 SE 26    SE 0.86 0.86 0.08 Table 1.2 (cont’d) Pitfall traps Drift fence Species ̂ SE ̂ RBSA 0.03 0.03 0.24 0.70 NSTS 0.04 0.14 0.48 Na Sorex spp. 0.12 0.02 0.86 Na a ̂ SE ̂ 0.61 0.12 0.52 0.81 Na b 0.14 0.03 0.62 0.74 0.14 b 0.24 0.03 0.95 0.99 0.05 SE b b Na Na SE b Na b We omitted parameter estimates for cover boards (all species) and VES (all species except red-backed salamander) because all detection probabilities were <0.15. b Na: unrealistic parameter estimate because of low detection probability ( ̂ < 0.05) or unsuccessful computation of the variance– covariance matrix for the given method. 27    Table 1.3. Ranking of candidate detection models for 3 wildlife species at 21 sites in northern Michigan, USA, 2009. Covariates include Julian date and 24-hr precipitation for each visit. A squared symbol represents the quadratic term for the respective covariate. I generated models using detection data from drift fences for northern short-tailed shrews and pygmy and masked shrews and visual encounter surveys for red-backed salamanders. All other method-species combinations yielded insufficient detection probabilities ( ̂  ≤ 0.15, see Results) for inclusion in this analysis. I report the complete model set for red-backed salamanders, but I report only models accounting for 0.90 of the total Akaike weight (wi) for the shrew species. 28    Table 1.3. (cont’d) AICc Species Model Northern short-tailed shrew Precipitation + precipitation a ΔAICc a wi a a K 2 122.4 0.0 0.80 4 2 125.5 3.1 0.17 5 2 234.3 0.0 0.64 5 Precipitation + Julian date + Julian date 236.5 2.2 0.21 5 Julian date 238.4 4.1 0.08 3 Julian date 83.4 b 0.0 0.54 3 Null model 85.4 b 2.0 0.20 2 b 3.1 0.12 4 b 4.5 0.06 3 b 5.9 0.03 4 b 6.2 0.02 5 b 6.2 0.02 5 b 10.2 0.00 6 Precipitation + precipitation + Julian date Pygmy and masked shrew Precipitation + precipitation + Julian date 2 Red-backed salamander Julian Date + Julian date 2 86.5 Precipitation 87.9 2 Precipitation + precipitation 89.3 2 Precipitation + Julian date + Julian date 89.6 2 Precipitation + precipitation + Julian date 2 89.6 2 Precipitation + precipitation + Julian date + Julian date 29    93.6 Table 1.3. (cont’d) a AICc = Akaike’s Information Criterion adjusted for small sample size; ΔAICc = difference from the best model: K = no. of parameters. b Evidence of poor model fit occurred for red-backed salamanders, therefore we used the quasi-likelihood version of AICc, QAICc for model ranking.   30    Table 1.4. Statistical power approximations for detecting a simulated change in northern short-tailed shrew and pygmy and masked shrew occupancy using 3 alternative sampling scenarios. For all scenarios I simulated detection history data for 2 groups of 25 sites that we visited 4 times in a season. I used the parameter estimates from the highest ranking model for drift fences (Table 3) to represent site occupancy probability for the first group of 25 sites ( 1 ) and for species detection probability ( ̂ ). To simulate a decline in species occupancy, I set the occupancy probability for the second group of sites ( 2 ) to 50% of 1 . Each scenario represents a situation where researchers target sampling effort during days within 24 hr of moderate (10 mm), light (5 mm), or no rainfall (0 mm). 31    Table 1.4. (cont’d) Power approximation (Ha vs. H0) Parameters for simulation 1 Power 2 (n = 25) (n = 25) ̂ χ 10mm rainfall 0.64 0.32 0.89 5mm rainfall 0.64 0.32 No rainfall 0.64 10mm rainfall 2 a Power a a df (α = 0.05) (α = 0.01) 5.2 1 63% 38% 0.61 5.0 1 61% 37% 0.32 0.03 0.0 1 5% 5% 0.99 0.49 0.44 12.4 1 94% 82% 5mm rainfall 0.99 0.49 0.31 7.9 1 80% 59% No rainfall 0.99 0.49 0.14 2.9 1 40% 19% Sampling scenario Northern short-tailed shrew Sorex spp. a Approximated power is the chance that the false null hypothesis (H0 = no difference in the occupancy probability of  1 and 2 is correctly rejected. 32    Figure 1.1. Schematic for a randomly selected 60 × 60-m site with 4 sampling methods to detect (species) in northern Michigan, USA, 2009. Background image represents a sampling lattice with 9 potential sampling sites and shaded areas denote retained canopy green-trees. Symbols represent an array of 4 19-L buckets (circles), 2 1.5 × 0.75-m plywood cover boards (squares), a 20-m visual encounter survey (line), or a drift fence with 5-m wings (sideways “Y”). Each drift fence had paired funnel traps (thin lines) and a bucket at each end and one in the center (circle). I spaced all sampling methods >10 m apart and >10 m from the site boundary. Visual encounter survey 10 m 60 m Coverboard array 60 m Drift fence Pitfall array 33    Figure 1.2. Estimated detection probability as a function of precipitation within a 24-hr period prior to sampling for (A) northern short-tailed shrews and (B) pygmy and masked shrews in northern Michigan, USA, 2009. I calculated detection probabilities using model averaged estimates based on the most supported models. A Detection probability (95% CI) 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 0 2 4 6 8 10 12 14 16 18 20 Precipitation (mm) B Detection probability (95% CI) 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 0 2 4 6 8 10 12 Precipitation (mm) 34    14 16 18 20 Figure 1.3. Estimated detection probability as a function of sampling date for (A) pygmy and masked shrews and (B) red-backed salamanders in northern Michigan, USA, 2009. I calculated detection probabilities using model averaged estimates based on the most supported models. A 1.0 Detection probability (95% CI) 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 Date B 1.0 Detection probability (95% CI) 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 Date 35    LITERATURE CITED 36    LITERATURE CITED Bailey, L. L., T. R. Simons, and K. H. Pollock. 2004a. Estimating detection probability parameters for plethodon salamanders using the robust capture-recapture design. Journal of Wildlife Management 68:1-13. Bailey, L. L., T. R. Simons, and K. H. Pollock. 2004b. Estimating site occupancy and species detection probability parameters for terrestrial salamanders. Ecological Applications 14:692-702. Bailey, L. L., J. E. Hines, J. D. Nichols, and D. I. MacKenzie. 2007. Sampling design trade-offs in occupancy studies with imperfect detection: examples and software. Ecological Applications 17:281-290. Barnes, B. V., and W. H. Wagner Jr. 2004. Michigan Trees: A Guide to the Trees of the Great Lakes Region University of Michigan Press, Ann Arbor, USA. Bielecki, J., J. Ferris, K. Kintigh, M. Koss, D. Kuhr, S. MacKinnon, S. Throop, L. Visser, and M. Walters. 2006. Within stand retention guidance. Michigan Department of Natural Resources and Environment, Lansing, USA. Brannon, M. P. 2002. Epigeal movement of the smoky shrew Sorex fumeus following precipitation in ridgetop and streamside habitats. Acta Theriologica 47:363-368. Burnham, K. P., and D. R. Anderson. 2002. Model selection and multi-model inference: a practical information-theoretic approach. Second edition. Springer-Verlag, New York, New York, USA. Burnham, K. P., D. R. Anderson, G. C. White, C. Brownie, and K. P. Pollock. 1987. Design and analysis of methods for fish survival experiments based on release–recapture. American Fisheries Society Monograph 5:1-437. Butts, S. R., and W. C. McComb. 2000. Associations of forest-floor vertebrates with coarse woody debris in managed forests of western Oregon. Journal of Wildlife Management 64:95-104. Corn, P. S. 1994. Straight-line drift fences and pitfall traps. Pages 109-117 in W. R. Heyer, M. A. Donnelly, R. W. McDiarmid, L. C. Hayek, and M. S. Foster, editors. Measuring and monitoring biological diversity: standard methods for amphibians. Smithsonian Institute Press, Washington, D.C., USA. 37    Crump, M. L., and N. J. Scott. 1994. Visual encounter surveys. Pages 84-92 in W. R. Heyer, M. A. Donnelly, R. W. McDiarmid, L. C. Hayek, and M. S. Foster, editors. Measuring and monitoring biological diversity: standard methods for amphibians. Smithsonian Institute Press, Washington, D.C., USA. Fellers, G. M., and C. A. Drost. 1994. Sampling with artificial cover. Pages 146-150 in W. R. Heyer, M. A. Donnelly, R. W. McDiarmid, L. C. Hayek, and M. S. Foster, editors. Measuring and monitoring biological diversity: standard methods for amphibians. Smithsonian Institute Press, Washington, D.C., USA. Ford, W. M., C. A. Dobony, and J. W. Edwards. 2002. Shrews in managed northern hardwood stands in the Allegheny mountains of West Virginia. Proceedings of the Annual Conference Southeastern Association of Fish and Wildlife Agencies 56:374-384. Garden, J. G., C. A. McAlpine, H. P. Possingham, and D. N. Jones. 2007. Using multiple survey methods to detect terrestrial reptiles and mammals: what are the most successful and cost-efficient combinations? Wildlife Research 34:218-227. Getz, L. L., and B. McGuire. 2008. Factors influencing movement distances and home ranges of the short-tailed shrew (Blarina brevicauda). Northeastern Naturalist 15:293-302. Gibbons, J. W., and R. D. Semlitsch. 1987. Activity patterns. Pages 396-421in R. A. Seigel, J. T. Collins, and S. S. Novak. Snakes: ecology and evolutionary biology. Macmillan, New York, New York, USA. Heatwole, H. 1962. Environmental factors influencing local distribution and activity of the salamander, Plethodon cinereus. Ecology 43:460-472. Hutchens, S. J., and C. S. DePerno. 2009. Efficacy of sampling techniques for determining species richness estimates of reptiles and amphibians. Wildlife Biology 15:113-122. Hyde, E. J., and T. R. Simons. 2001. Sampling plethodontid salamanders: sources of variability. Journal of Wildlife Management 65:624-632. Jackson, J. T., F. W. Weckerly, T. M. Swannack, and M. R. Forstner. 2006. Inferring absence of Houston Toads given imperfect detection probabilities. Journal of Wildlife Management 70:1461-1463. Kéry, M., and B. R. Schmidt. 2008. Imperfect detection and its consequences for monitoring for conservation. Community Ecology 9:207-216. Kleeberger, S. R., and J. K. Werner. 1982. Home range and homing behavior of Plethodoncinereus in northern Michigan. Copeia 1982:409-415. 38    Kroll, A. J. 2009. Sources of uncertainty in stream-associated amphibian ecology and responses to forest management in the Pacific Northwest, USA: A review. Forest Ecology and Management 257:1188-1199. Kurta, A. 1995. Mammals of the great lakes region. The University of Michigan Press, Ann Arbor, USA. Laakkonen, J., R. N. Fisher, and T. J. Case. 2003. Factors influencing the variation in capture rates of shrews in southern California, USA. Acta Theriologica 48:157-1666. Marsh, D. M., and M. A. Goicochea. 2003. Monitoring Terrestrial Salamanders: Biases Caused by Intense Sampling and Choice of Cover Objects. Journal of Herpetology 37:460-466. MacKenzie, D. I. 2005. What are the issues with presence-absence data for wildlife managers? Journal of Wildlife Management 69:849-860. MacKenzie, D. I., and L. L. Bailey. 2004. Assessing the fit of site-occupancy models. Journal of Agricultural, Biological, and Environmental Statistics 9:300-318. MacKenzie, D. I., and J. A. Royle. 2005. Designing occupancy studies: general advice and allocating survey effort. Journal of Applied Ecology 42:1105-1114. MacKenzie, D. I., J. D. Nichols, G. B. Lachman, S. Droege, J. A. Royle, and C. A. Langtimm. 2002. Estimating site occupancy rates when detection probabilities are less than one. Ecology 83:2248-2255. MacKenzie, D. I., J. D. Nichols, N. Sutton, K. Kawanishi, and L. L. Bailey. 2005. Improving inferences in population studies of rare species that are detected imperfectly. Ecology 86:1101-1113. MacKenzie, D.I., J. D. Nichols, J. A. Royle, K. H. Pollock, L. L. Bailey, and J. E. Hines. 2006. Occupancy estimation and modeling: inferring patterns and dynamics of species occurrence. Academic Press, New York, New York, USA. Marsh, D. M., and M. A. Goicochea. 2003. Monitoring terrestrial salamanders: biases caused by intense sampling and choice of cover objects. Journal of Herpetology 37:460-466. Mattfeldt, S.D., and E. H. C. Grant. 2007. Are two methods better than one?: area constrained transects and leaf litterbags for sampling stream salamanders. Herpetological Review 38:43-45. Mattfeldt, S. D., L. L. Bailey, and E. H. Grant. 2009. Monitoring multiple species: Estimating state variables and exploring the efficacy of a monitoring program. Biological Conservation 142:720-737. 39    Mazerolle, M. J., A. Desrochers, and L. Rochefort. 2005. Landscape characteristics influence pond occupancy by frogs after accounting for detectability. Ecological Applications 15:824-834. Mazerolle, M. J., L. L. Bailey, W. L. Kendall, J. A. Royle, S. J. Converse, and J. D. Nichols. 2007. Making great leaps forward: Accounting for detectability in herpetological field studies. Journal of Herpetology 41:672-689. McDade, K. A., and C. C. Maguire. 2005. Comparative effectiveness of three techniques for salamander and gastropod land surveys. American Midland Naturalist 153:309-320. Michigan Department of Information Technology 2007. 2005 National agricultural imagery program data (NAIP 2005). . Accessed 01 Feb 2009). Monti, L., M. Hunter, and J. Witham. 2000. An evaluation of the artificial cover object (ACO) method for monitoring populations of the redback salamander Plethodon cinereus. Journal of Herpetology 34:624-629. Nichols, J. D., L. L. Bailey, A. F. O'Connell, N. W. Talancy, E. H. Grant, A. T. Gilbert, E. M. Annand, T. P. Husband, and J. E. Hines. 2008. Multi-scale occupancy estimation and modelling using multiple detection methods. Journal of Applied Ecology 45:1321-1329. O'Connell, A. F., N. W. Talancy, L. L. Bailey, J. R. Sauer, R. Cook, and A. T. Gilbert. 2006. Estimating site occupancy and detection probability parameters for meso- and large mammals in a coastal ecosystem. Journal of Wildlife Management 70:1625-1633. Petranka, J. W. 1998. Salamanders of the United States and Canada. The Smithsonian Institution, Washington D. C., USA. Prugh, L. R., K. E. Hodges, A. R. Sinclair, and J. S. Brashares. 2008. Effect of habitat area and isolation on fragmented animal populations. Proceedings of the National Academy of Sciences of the United States of America 105:20770-20775. Ribeiro-Júnior, M. A., T. A. Gardner, and T. C. Ávila-Pires. 2008. Evaluating the Effectiveness of Herpetofaunal Sampling Techniques across a Gradient of Habitat Change in a Tropical Forest Landscape. Journal of Herpetology 42:733-749. Rittenhouse, T. A., and R. D. Semlitsch. 2007. Distribution of amphibians in terrestrial habitat surrounding wetlands. Wetlands 27:153-161. Rittenhouse, T. A., R. D. Semlitsch, and F. R. Thompson. 2009. Survival costs associated with wood frog breeding migrations: effects of timber harvest and drought. Ecology 90:16201630. 40    Roloff, G. J., T. E. Grazia, K. F. Millenbah, and A. J. Kroll. In press. Factors associated with amphibian detection and occupancy in southern Michigan forests. Journal of Herpetology. Ryan, T. J., T. Philippi, Y. A. Leiden, M. E. Dorcas, T. B. Wigley, and J. W. Gibbons. 2002. Monitoring herpetofauna in a managed forest landscape: effects of habitat types and census techniques. Forest Ecology and Management 167:83-90. Semlitsch, R. D., and R. Bodie. 2003. Biological criteria for buffer zones around wetlands and riparian habitats for amphibians and reptiles. Conservation Biology 17:1219-1228. Smith, C. K., and J. W. Petranka. 2000. Monitoring terrestrial salamanders: Repeatability and validity of area-constrained cover object searches. Journal of Herpetology 34:547-557. Sokal, R. R., and F. J. Rohlf. 1981. Biometry. Second Edition. W. H. Freeman and Company, New York, New York, USA. Umetsu, F., L. Naxara, and R. Pardini. 2006. Evaluating the efficiency of pitfall traps for sampling small mammals in the neotropics. Journal of Mammalogy 87:757-765. Whittaker, J. C., and G. A. Feldhamer. 2005. Population dynamics and activity of southern shorttailed shrews (Blarina carolinensis) in southern Illinois. Journal of Mammalogy 86:294301. Yoccoz, N. G., J. D. Nichols, and T. Boulinier. 2001. Monitoring of biological diversity in space and time. Trends in Ecology and Evolution 16:446-453. 41    CHAPTER 2 COMPARING COVER OBJECT AND LEAF LITTER SURVEYS FOR DETECTING RED-BACKED SALAMANDERS, PLETHODON CINEREUS Abstract Sampling methods that can be efficiently implemented while minimizing bias are central for achieving strong statistical power in field studies. I estimated site occupancy and detection probabilities for Red-backed Salamanders, Plethodon cinereus, using natural cover object (NCO) and leaf litter surveys, and tested the power of these methods to detect occupancy differences in simulated data. Both NCO and leaf litter surveys performed well in their ability to detect P. cinereus (  = 0.59 ± 0.07 (±1SE); 0.49 ± 0.08, respectively) and produced similar site occupancy estimates ( = 0.75 ± 0.09; 0.77 ± 0.11, respectively); however, leaf litter searches took 3 times longer to complete. My simulation study revealed the time required to complete leaf litter surveys limited the number of sampling sites, which reduced statistical power for detecting a 25% change in salamander occupancy between two groups of sites. Natural cover object searches provided greater power for detecting a similar change in occupancy, largely because of high sampling size. Although reliable salamander detection data can be collected quickly using NCO surveys, I discuss situations where this technique may not be appropriate. My study highlights the importance of accounting for imperfect detection probability when conducting visual encounter surveys for terrestrial plethodontids. Status: Published. Otto, C.R.V., Roloff, G.J., 2011. Comparing Cover Object and Leaf Litter Surveys for Detecting Red-Backed Salamanders, Plethodon cinereus. Journal of Herpetology 45, 256-260. 2.1 Introduction 42    Terrestrial salamanders of the genus Plethodon are important components of forest ecosystems in eastern North America through their contribution to forest biomass and as potential indicators of environmental change (Burton and Likens, 1975; Welsh and Droege, 2001; Welsh and Hodgson, 2008). As a result, this genus has been utilized in many ecological and forest management studies (Petranka et al., 1993; Ash, 1997; Davic and Welsh, 2004; Bailey et al. 2004a). Visual encounter surveys (VES) are commonly used for sampling terrestrial plethodontids because they are relatively easy to conduct and produce high detection probabilities and capture rates (Bailey et al., 2004a; Williams and Berkson, 2004; Strain et al., 2009). Most VES focus on detecting plethodontids under natural cover objects (NCO: woody debris, logs, rocks). When making inference, researchers assume the number of salamanders detected under these objects is positively correlated with true population size. However, research has shown that a substantial proportion of salamanders occur in the leaf litter or underground and therefore are unavailable for sampling under NCOs (Petranka, 1998:341-342). This suggests the use of NCO surveys alone could lead to biased population metrics if researchers do not account for imperfect detection probability or employ more rigorous sampling methods. Strain et al. (2009) compared NCO and leaf litter surveys using stream salamander counts and species richness as their metric of inference. They found that NCO surveys were generally more cost effective than leaf litter searches, though their analysis did not account for imperfect detection. Leaf litter searches have the potential to yield more accurate population information but require more time to complete. If only a small number of sites can be visited in a field season, inferential power may be limited. Comparing detection estimates and statistical power of these two methods would provide valuable information to researchers developing long-term monitoring programs or forest management studies (Field et al., 2005). 43    I compared the effectiveness of NCO and leaf litter searches by estimating occupancy and detection probability parameters for Red-backed Salamanders (Plethodon cinereus) occurring in harvested forests. Because a substantial proportion of salamanders may reside in the leaf litter (Petranka, 1998:341-342; Bailey et al., 2004b), I predicted that NCO surveys would yield lower detection probabilities and hence overestimate salamander occupancy with reduced precision. Furthermore, I predicted an analysis that combines NCO and leaf litter searches would yield the most precise estimates of detection and occupancy. I tested whether survey duration and the number of search NCOs influenced P. cinereus detection probability. Last, I conducted a simulation to test the statistical power of each sampling method for detecting a 25% difference in salamander occupancy between 2 site types. 2.2. Methods 2.2.1. Site Selection My study occurred in the northwestern Lower Peninsula of Michigan, USA, in 2009. I sampled 6, stand-owned, forest stands that are currently managed for aspen production (Populus sp.). The dominant tree species on these stands include quaking (Populus tremuloides) and big-tooth aspen (P. grandidentata) interspersed with eastern white pine (Pinus strobus), red maple (Acer rubrum) and mixed oak (Quecus sp.). Within each stand, I randomly selected 5 sites, yielding a total of 30 sites, where I searched for salamanders. To minimize the influence of habitat heterogeneity on salamander occupancy I did not selects sites that were within 30m of a forest stand boundary, active logging road, off-road recreational vehicle trail or isolated wetland. Each site consisted of a clearly demarcated 10 x 10m square. I divided each square into three 3.3 x 10m sub-plots where a single observer would perform NCO and leaf litter searches to detect P. cinereus. 2.2.2. Salamander Searches 44    Observers began searches at one end of the sub-plot and worked lengthwise to the opposite end. For the initial survey an observer walked through each sub-plot and searched for P. cinereus under and within visible NCOs that were >4 cm in diameter. Rotting logs that readily broke apart were thoroughly searched. During this survey all NCOs were removed from the sub-plot before starting leaf litter searches. To minimize pseudo-replication, observers also removed all detected salamanders from the sub-plot immediately after capture. Each observer recorded the number of searched NCOs and survey duration. After finishing the NCO survey within their assigned sub-plot the same observer performed a systematic search through all detritus, including leaf litter, coarse organic matter, and NCOs that were missed during the first survey. As before, all salamanders were removed after capture and search time and the number of newly discovered NCOs were recorded. 2.2.3. Statistical Analyses I used single-season occupancy models (MacKenzie et al., 2002) and the program PRESENCE (version 2; James E. Hines, Patuxent Wildlife Research Center, Laurel, MD) to estimate occupancy and detection probability parameters for P. cinereus. I generated a detection history at each site by treating the three sub-plots as spatial subunits (MacKenzie et al., 2006:161-162; Kendall and White, 2009). I initially combined detection information for NCO and leaf litter surveys as a single sampling event (‘combined’ dataset) for each site. These data represent the likelihood that P. cinereus was detected using either a NCO or leaf litter survey at one or more sub-plots, given its presence at a site. Because this dataset includes detection information from both methods, it should yield the most accurate estimate of the true occupancy state at a site and the highest detection probability. 45    For my next analysis I created unique datasets for the NCO and leaf litter searches and analyzed each separately (‘single-method’ datasets). This represents a situation where only one method was used at a site, hence the potential for bias in site occupancy estimation was higher. Comparing site occupancy estimates between the ‘single-method’ and ‘combined’ datasets allowed me to investigate occupancy biases associated with each method, as well as differences in method-specific detection probability (Bailey et al., 2004a; Mattfeldt and Grant, 2007). Occupancy estimates with non-overlapping confidence intervals generated from these two analyses would suggest a bias associated with one of the sampling methods. I also developed a set of a priori models to better understand how P. cinereus detection was influenced by survey effort. I modeled detection probability as a function of two covariates: the number of NCOs searched (COVER) and survey duration (TIME), while holding occupancy constant across all sites. I tested for correlation between COVER and TIME for both survey methods using Pearson’s product moment correlation (Sokal and Rohlf 1981). I did not record covariate information at 2 and 3 sites for the NCO and leaf litter surveys, respectively. For these sites, I entered the covariates as missing values in program PRESENCE. I standardized all covariates prior to analysis. Akaike’s Information Criterion, adjusted for small sample size (AICc), and model weights were used to rank models in order of parsimony (Burnham and Anderson, 2002). For the NCO dataset I predicted that the COVER model would receive the strongest support, with detection positively associated with number of NCOs. For the leaf litter dataset I predicted that the TIME model would receive the strongest support, with salamander detection increasing with survey duration. I assumed the probability of detecting P. cinereus during a leaf litter search was independent of salamanders removed during the NCO searches. To test this I constructed a 46    model where salamander detection probability during a leaf litter survey was allowed to vary as a function of whether at least one P. cinereus was removed from the sub-plot during the initial NCO survey (REMOVAL). A negative beta parameter for the REMOVAL covariate would suggest that salamander detection probability during a leaf litter survey was lower at sub-plots where salamanders were removed during the initial NCO survey. This would provide evidence for a lack of method independence. I also created a single dataset where detection information was included separately for each sampling method (‘multi-method’), yielding a total of 6 sampling replicates (3 sub-plots x 2 sampling methods) for each site. I constructed two models where detection was either held constant (NULL) or allowed to vary by sampling method (METHOD). Support for the model including the METHOD covariate would suggest survey methods differ in their ability to detect P. cinereus. 2.2.4. Power Approximation I conducted a simulation analysis to investigate how use of these two methods would affect statistical power of future study designs, with the understanding that sample size in field studies is often limited by time availability and project funding. My goal was to determine how survey method affected my ability to detect a 25% change in the proportion of sites occupied by P. cinereus. I utilized field data to estimate the maximum number of 10 x 10m sites that could realistically be surveyed in a 2-month field season using either method ( = 180 for NCO and 60 for leaf litter surveys). I then simulated salamander detection data for two groups of sites in the program GENPRES (James E. Hines, Patuxent Wildlife Research Center, Laurel, MD) using occupancy and detection probability estimates from my previous analysis (Table 1). I assigned the first group of sites a 0.78 occupancy probability (see Table 1) and calculated a 25% decrease 47    in occupancy for the second group ( 2 = 0.59). The estimated number of sampling sites was split evenly between each group. Using the Program MARK (White and Burnham, 1999), I compared a null model (H0) that assumed occupancy was constant between the 2 groups to an alternative model (Ha) that allowed occupancy to vary between groups. I considered Ha as the true model, as my simulated data were generated using a separate occupancy estimate for the two groups. I compared both models using a likelihood ratio test (LRT) in MARK (White and Burnham, 1999) and approximated power (Burnham et al., 1987:214-217; Bailey et al., 2007) using the chi-square statistic from the LRT output. 2.3. Results Natural cover object and leaf litter surveys took an average of 16.3 ± 0.9(SE) and 42.6 ± 1.6 minutes to complete at each sub-plot, respectively. I searched through an average of 21.1 ± 1.2 and 10.9 ± 1.2 NCOs during the NCO and leaf litter surveys, respectively. Natural cover object searches yielded slightly higher detection estimates than leaf litter searches; however, both methods had overlapping 95% confidence intervals (95% CI; Table 1). For all analyses naïve occupancy (proportion of sites known to be occupied, without accounting for imperfect detection) was smaller than occupancy estimates adjusted for imperfect detection (Table 1). Site occupancy estimates from the ‘single-method’ and ‘combined’ datasets had overlapping 95% CI, suggesting low bias in occupancy estimation for either method (Table 1). Modeling results were counter to my a priori predictions for both method-specific analyses. For the NCO analysis, the detection model containing the COVER covariate received virtually no support when compared to the model including the TIME covariate (Table 2). The positive beta parameter estimate for TIME indicated that salamander detection probability increased with survey duration ( = 1.46 ± 0.42; Figure 2.1). For the leaf litter analysis, the 48    model containing COVER received the strongest support, with salamander detection probability decreasing as the number of searched NCOs increased ( = −0.77 ± 0.39; Figure 2.2). I found little support for the amount of search time influencing salamander detection probability during leaf litter surveys (ΔAICc = 6.56, wi = 0.04; Table 2). I found no evidence that removal of salamanders during the NCO searches negatively impacted my ability to detect salamanders during the leaf litter searches (ΔAICc = 17.06, wi = 0.00). The beta parameter for the REMOVAL covariate was positive, suggesting I was more likely to detect P. cinereus during a leaf litter survey if I removed at least one salamander during the NCO survey ( = 0.80 ± 0.56). COVER and TIME were weakly correlated for the NCO surveys (r = 0.31, t82 = 2.96, P = 0.004), but not for leaf litter surveys (r = −0.06, t79 = 0.05, P = 0.61). Ranking for the ‘multi-method’ analysis indicated model selection uncertainty between the NULL and METHOD detection models (Table 2). Although NULL was my top supported model, the second best model, METHOD, received 32% of the total model weight and was <2 AICc from the top model (Table 2). However, estimated detection probabilities for both sampling methods in the METHOD model overlapped considerably (95% CINCO = 0.45-0.69, and 95% CILeaf litter = 0.37-0.61). This finding, combined with results from the individual NCO and leaf litter survey analyses, suggest NCO and leaf litter searches yielded comparable occupancy and detection estimates for P. cinereus. There was substantial variation in the ability of either sampling method to detect a 25% change in salamander site occupancy (Table 3). Approximated power values for NCO and leaf litter surveys were 76% and 5%, respectively (α = 0.05). 49    2.4. Discussion My results show that NCO searches are a more cost-effective method than leaf litter searches for detecting P. cinereus in harvested aspen stands. Counter to my prediction, the use of two sampling methods only slightly improved the precision of occupancy estimates for P. cinereus. Both methods performed well in their ability to detect P. cinereus; however, my field crew and I could conduct ≈3 NCO surveys in the time required to complete one leaf litter survey. Hence, leaf litter searches merely provided redundant detection information at a substantial time cost. Because most herpetological field studies operate on a finite time budget and limited sampling size, it is important to understand how sampling method(s) may affect statistical power, and ultimately scientific inference (Smith and Petranka, 2000; Bailey et al., 2007). My simulation study revealed that leaf litter searches yielded surprisingly low power for detecting a 25% change in salamander occupancy between two groups of sites, largely because of low sample size. Hence, even though leaf litter searches may generate reliable occupancy and detection estimates, the time required to conduct them limited the number of sites that could be visited during a simulated field season, which increased my probability of making a Type II Error (Sokal and Rolf, 1981:159-169). The higher approximated power for NCO surveys was because of the large number of sampling sites that could be visited, while still providing relatively high P. cinereus detection probabilities. Counter to my prediction, I found little support for the model testing the influence of the number of searched cover objects on detecting P. cinereus during NCO surveys. Coarse woody debris provides ideal micro-habitat conditions for P. cinereus and is an important structural component for recovering salamander populations in harvested forests (e.g., McKenny et al., 2006, Owens et al., 2008). Therefore, I predicted that searching more cover objects should yield 50    a higher probability of detection. Rather, model selection statistics revealed strongest support for the NCO survey model that included survey duration on salamander detection. For my study the number of searched NCOs ranged from 15 to 112, which may be above the minimum number of cover objects required to affect detection probability. Furthermore, large, well-decayed logs required considerable time to search through, but only counted as single NCO in my analysis. This potentially explains why my TIME model received more support than COVER. Future analyses should explore how NCO size and decay-class influences salamander detection. Counter to my predictions, the leaf litter analysis revealed P. cinereus detection was 2 negatively correlated with the number of searched NCOs. Some of my sites contained 1 to 2m piles of woody debris fragments (i.e., slash piles), which may explain the high number of NCOs found during the leaf litter surveys at these sites. These slash piles were difficult to search through and likely hindered my ability to detect salamanders. I found little support for the model containing survey duration on salamander detection probability during the leaf litter surveys. At this time it is not clear why spending additional search time may improve P. cinereus detection during NCO surveys, but not during leaf litter surveys. My generated parameter estimates are specific to the methodologies I used and are subject to change depending on sampling protocol or study organism. For example, in my study observers conducting NCO searches thoroughly dismantled all rotting logs, which likely resulted in higher detection probabilities than if observers rolled them over. Dismantling logs may not be appropriate for other studies where sites need to be revisited or when dealing with species of conservation concern (e.g., Flint and Harris, 2005; O’Donnell et al., 2009). The invasiveness of NCO searches in previous terrestrial salamander studies varies from totally dismantling cover objects (O’Donnell et al., 2009; Kluber et al., 2008; this study) to merely turning them over 51    (Bailey et al., 2004a; McKenny et al., 2006; Strain et al., 2009). At this time it is unclear how the rigor of NCO searches may impact salamander population or detection estimates. Furthermore, my results show that some P. cinereus reside in the leaf litter, hence conducting only NCO surveys may lead to an incorrect characterization of site occupancy at sites lacking NCOs. This is an important point of consideration for research conducted in areas with low numbers of NCOs. Time investment and statistical power are two factors that researchers must consider when selecting a method(s) for herpetological research (Smith and Petranka, 2000; Mattfeldt et al., 2009; Strain et al., 2009). I found NCO surveys were a more cost-effective method than leaf litter searches for estimating P. cinereus site occupancy and detection probabilities and, as a consequence, yielded greater statistical power. For both of my methods, detection probability was consistently <1, which highlights the importance of accounting for imperfect detection. Whether using the proportion of sites occupied or abundance as the state variable, herpetologists should consider how imperfect detection probability may bias results (Mazerolle et al., 2007). Furthermore, models developed by MacKenzie et al. (2006) and Nichols et al. (2008) provide a flexible framework for addressing potential biases associated with sampling methodology. 2.5. Acknowledgements I thank A. Coleman and R. Sterling for aiding in the collection of field data and M. Donovan and K. Fitzpatrick with developing project objectives. I thank G. Perry and 2 anonymous reviewers for strengthening this manuscript. Funding for this research was provided by the Michigan Department of Natural Resources and Environment Wildlife Division with funds from the federal Pittman-Robertson Wildlife Restoration Act grant administered by the US Fish and Wildlife (W-147-R: Michigan’s Statewide Wildlife Research and Restoration Program). All 52    sampling and handling protocol used in this study were approved by the Michigan State University Animal Care and Use Committee (AUF# 07/08-118-00). 53    Table 2.1. Parameter estimates for occupancy ( ) and detection probabilities ( ̂ ) for Red-backed Salamanders using natural cover object (NCO) and leaf litter surveys at 30 sites in northern Michigan, 2009 (95% confidence interval in parentheses). (obs) is the proportion of sites occupied without accounting for imperfect detection. All parameters were estimated from a model assuming constant occupancy and detection probabilities. Survey method ̂ (obs) NCO survey 0.59 (0.45-0.72) 0.70 0.75 (0.53-0.89) Leaf litter survey 0.49 (0.34-0.65) 0.67 0.77 (0.48-0.91) Combined methods 0.77 (0.65-0.86) 0.77 0.78 (0.59-0.89) 54    Table 2.2. Ranking of candidate models for Red-backed Salamanders detected using natural cover object (NCO) and leaf litter surveys at 30 sites in northern Michigan, 2009. Detection covariates include the number of searched NCOs (COVER), survey duration (TIME) and survey type (METHOD). The NULL model assumes detection probability was constant across all subplots. ‘Multi-method’ includes data from both surveys types. For all models, occupancy was held constant between sites. Weight (wi) and K represent the probability that model i is the best in the candidate set and number of estimated parameters, respectively. Logit-link parameter estimates are provided for the detection portion of each model ( 0 = detection intercept estimate, 1 = detection covariate estimate). 55    Table 2.2. (cont’d) AICc ΔAICc wi K 0 (SE) 1 (SE) TIME 98.26 0.00 1.00 3 0.13 (0.36) 1.46 (0.42) COVER 112.84 14.58 0.00 3 0.41 (0.32) 0.33 (0.31) NULL 121.32 23.06 0.00 2 0.37 (0.30) - COVER 104.14 0.00 0.96 3 −0.60 (0.42) −0.77 (0.39) TIME 110.73 6.59 0.04 3 −0.51 (0.39) 0.61 (0.29) NULL 120.46 16.32 0.00 2 −0.03 (0.32) - NULL 227.15 0.00 0.67 2 0.12 (0.18) - METHOD 228.59 1.44 0.33 3 −0.05 (0.24) 0.35 (0.34) Model NCO surveys Leaf litter surveys Multi-method 56    Table 2.3. Parameter estimates used to simulate a 25% change in salamander occupancy probability and power approximations (Burnham et al., 1987:214-217; Bailey et al., 2007) for detecting the simulated change between two groups of sites using natural cover object (NCO) or leaf litter surveys. Occupancy ( ) and detection probability ( ̂ ) estimates were generated from Table 1, while the maximum sample size ( ) was estimated from my field data. Approximated power is the chance that the false null hypothesis (i.e., no difference in the occupancy probability of g1 and g2) was correctly rejected. Power approximation (Ha vs H0) Parameters used for simulation Sampling method χ ̂ 2 df Power (α = 0.05) Power (α = 0.01) g1 g2 NCO survey 0.78 0.59 0.59 180 6.2 1 70% 47% Leaf litter survey 0.78 0.59 0.49 60 0.0 1 5% 1% 57    Figure 2.1. Estimated detection probability for Red-backed salamanders modeled as a function of survey duration during natural cover object surveys in northern Michigan, USA, 2009. 1.0 0.9 Detection Probability (95% CI) 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 5 10 15 20 25 30 35 Survey Duration (minutes) 58    40 45 50   Figure 2.2. Estimated detection probability for Red-backed Salamanders modeled as a function of the number of cover objects that were searched through during leaf litter surveys in northern Michigan, USA, 2009. 1.0 0.9 Detection Probability (95% CI) 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 0 5 10 15 20 25 30 Number of Cover Objects 59    35 40 45   LITERATURE CITED 60      LITERATURE CITED Ash, A. N. 1997. Disappearance and return of plethodontid salamanders to clearcut plots in the Southern Blue Ridge Mountains. Conservation Biology 11:983-989. Bailey, L. L., T. R. Simons, and K. H. Pollock. 2004a. Estimating site occupancy and species detection probability parameters for terrestrial salamanders. Ecological Applications 14:692-702. Bailey, L. L., T. R. Simons, and K. H. Pollock. 2004b. Estimating detection probability parameters for plethodon salamanders using the robust capture-recapture design. Journal of Wildlife Management 68:1-13. Bailey, L. L., J. E. Hines, J. D. Nichols, AND D. I. MacKenzie. 2007. Sampling design trade-offs in occupancy studies with imperfect detection: Examples and software. Ecological Applications 17:281-290. Burnham, K. P., AND D. R. Anderson. 2002. Model Selection and Multi-model Inference: a Practical Information-theoretic Approach. Springer, New York, USA. Burnham, K. P., D. R. Anderson, G. C. White, C. Brownie, and K. P. Pollock. 1987. Design and analysis of methods for fish survival experiments based on release–recapture. American Fisheries Society Monograph 5:1–437. Burton, T. M., and G. E. Likens. 1975. Salamander populations and biomass in Hubbard Brook Experimental Forest, New-Hampshire. Copeia 1971:541-546. Davic, R. D., and H. H. Welsh. 2004. On the ecological roles of salamanders. Annual Review of Ecology Evolution and Systematics 35:405-434. Field, S. A., A. J. Tyre, and H. P. Possingham. 2005. Optimizing allocation of monitoring effort under economic and observational constraints. Journal of Wildlife Management 69:473482. Flint, W. D., and R. N. Harris. 2005. The efficacy of visual encounter surveys for population monitoring of Plethodon punctatus (Caudata : Plethodontidae). Journal of Herpetology 39:578-584. Kendall, W. L., and G. C. White. 2009. A cautionary note on substituting spatial subunits for repeated temporal sampling in studies of site occupancy. Journal of Applied Ecology 46:1182-1188. Kluber, M. R., D. H. Olson, and K. J. Puettmann. 2008. Amphibian distributions in riparian and upslope areas and their habitat associations on managed forest landscapes in the Oregon Coast Range. Forest Ecology and Management 256:529-535. 61      MacKenzie, D. I., J. D. Nichols, G. B. Lachman, S. Droege, J. A. Royle, and C. A. Langtimm. 2002. Estimating site occupancy rates when detection probabilities are less than one. Ecology 83:2248-2255. MacKenzie, D.I., J. D. Nichols, J. A. Royle, K. H. Pollock, L. L. Bailey, and J. E. Hines. 2006. Occupancy Estimation and Modeling: Inferring Patterns and Dynamics of Species Occurrence. Academic Press, New York, USA. Mazerolle, M. J., L. L. Bailey, W. L. Kendall, J. A. Royle, S. J. Converse, and J. D. Nichols. 2007. Making great leaps forward: Accounting for detectability in herpetological field studies. Journal of Herpetology 41:672-689. McKenny, H. C., W. S. Keeton, and T. M. Donovan. 2006. Effects of structural complexity enhancement on eastern Red-backed Salamander (Plethodon cinereus) populations in northern hardwood forests. Forest Ecology and Management 230:186-196. Mattfeldt, S.D., and E. H. C. Grant. 2007. Are two methods better than one?: area constrained transects and leaf litterbags for sampling stream salamanders. Herpetological Review 38:43-45. Mattfeldt, S. D., L. L. Bailey, AND E. H. C. Grant. 2009. Monitoring multiple species: Estimating state variables and exploring the efficacy of a monitoring program. Biological Conservation 142:720-737. Nichols, J. D., L. L. Bailey, A. F. O'Connell, N. W. Talancy, E. H. C. Grant, A. T. Gilbert, E. M. Annand, T. P. Husband, and J. E. Hines. 2008. Multi-scale occupancy estimation and modelling using multiple detection methods. Journal of Applied Ecology 45:1321-1329. O'Donnell, R. P., T. Quinn, M. P. Hayes, and K. E. Ryding. 2009. Comparison of Three Methods for Surveying Amphibians in Forested Seep Habitats in Washington State. Northwest Science 81:274-283. Owens, A. K., K. R. Moseley, T. S. McCay, S. B. Castleberry, J. C. Kilgo, and W. M. Ford. 2008. Amphibian and reptile community response to coarse woody debris manipulations in upland loblolly pine (Pinus taeda) forests. Forest Ecology and Management 256:20782083. Petranka, J. W. 1998. Salamanders of the United States and Canada. Smithsonian Institution Press, Washington D. C., USA. Petranka, J. W., M. E. Eldridge, and K. E. Haley. 1993. Effects of timber harvesting on southern Appalachian salamanders. Conservation Biology 7:363-377. Smith, C. K., and J. W. Petranka. 2000. Monitoring terrestrial salamanders: Repeatability and validity of area-constrained cover object searches. Journal of Herpetology 34:547-557. Sokal, R. R., and F. J. Rohlf. 1981. Biometry. Second Edition. W. H. Freeman and Company, New York. 62      Strain, G. F., R. L. Raesly, AND R. H. Hilderbrand. 2009. A comparison of techniques to sample salamander assemblages along highland streams of Maryland. Environmental Monitoring and Assessment 156:1-16. Welsh, H. H., and S. Droege. 2001. A case for using plethodontid salamanders for monitoring biodiversity and ecosystem integrity of North American forests. Conservation Biology 15:558-569. Welsh, H. H., and G. R. Hodgson. 2008. Amphibians as metrics of critical biological thresholds in forested headwater streams of the Pacific Northwest, USA. Freshwater Biology 53:1470-1488. White, G. C., and K. P. Burnham. 1999. Program MARK: survival estimation from populations of marked animals. Bird Study 46 (Supplement):120–138. Willams, A. K., and J. Berkson. 2004. Reducing false absences in survey data: Detection probabilities of Red-backed Salamanders. Journal of Wildlife Management 68:418-428. 63      CHAPTER 3 IMPROVING OCCUPANCY ESTIMATION WHEN SAMPLING DISTURBANCES AND ANIMAL MOVEMENTS VIOLATE THE CLOSURE ASSUMPTION Abstract Site occupancy models that account for species imperfect detection are increasingly utilized in ecological research and wildlife monitoring. Occupancy models require replicate surveys to estimate detection probability over a time period where the occupancy status at sampled sites is assumed closed (i.e., no changes in occupancy due to local extinction or colonization). Unlike mark-recapture models, few studies have examined how violations of closure can bias occupancy estimates. My study design allowed me to differentiate among two processes that violate the closure assumption during a survey season: 1) repeated destructive sampling events that result in either short- or long-term site avoidance by the target species and 2) sampling occurring over a time period during which non-random movements of the target species result in variable occupancy status. I used dynamic occupancy models to quantify the potential bias in occupancy estimation associated with these processes for a terrestrial salamander system. I sampled 126 transects repeatedly within a single field season using invasive sampling techniques and constructed 17 candidate models that portrayed different sources of variation in occupancy, extinction, and detection probabilities. Model ranking indicated that both repeated sampling and non-random movement of organisms were evident in my system. I observed a chronic decrease in occupancy associated with cumulative sampling and non-random salamander movements over the entire field season. I also observed a strong but temporary disturbance effect on salamander detection probability associated with repeated sampling within a 24-hr. period. I generalized my findings by conducting a simulation to evaluate how violations of closure can bias occupancy 64      estimates when organism availability declines within a field season (i.e., when local extinction occurs within the season). My simulation study revealed general sensitivity of estimates from single-season occupancy models to violations of closure, with the strength and direction of bias varying between scenarios. Bias was minimal when extinction probability or the number of sample occasions was relatively low. My research highlights the importance of addressing closure in occupancy studies and I provide multiple solutions, using both design- and modelbased frameworks, for minimizing bias associated with non-random changes in occupancy and repeated sampling disturbances. Keywords: amphibian, detection probability, dynamic occupancy, extinction, multi-season, Plethodon cinereus, salamander, sampling bias, simulation. Status: This manuscript has successfully made it through the first round of reviews in Ecography. Revisions are ongoing. 3.1 Introduction Site occupancy models that account for species imperfect detection are increasingly utilized in ecological research and conservation planning (Marsh and Trenham 2008). For example, occupancy models have been used to assess the effects of anthropogenic disturbances on species distributions (Ferraz et al. 2007, Kroll et al. 2008, Zuckerberg et al. 2011), determine the efficacy of wildlife monitoring programs (Mattfeldt et al. 2009, Collier et al. 2010, Weller 2008), and investigate species habitat selection (Seamans and Gutierrez 2007). Occupancy models are generally preferred over traditional logistic regression or incidence function models because of their ability to account for false-absences caused by imperfect species detection (Mazerolle et al. 2005, MacKenzie et al. 2006). Although a number of occupancy models have been developed to account for imperfect detection (e.g., Giessler and Fuller 1987, MacKenzie et al. 2002, Nichols 65      and Karanth 2002, Tyre et al. 2003), the model developed by MacKenzie et al. (2002) has been the most readily adopted and numerous extensions of this model are available for use (reviewed in MacKenzie et al. 2006, Nichols et al. 2008, Martin et al. 2010). To account for false absences, the MacKenzie et al. (2002) occupancy model uses detection and non-detection data collected during multiple site surveys to estimate species detection probability. These repeated surveys are typically performed within a time period where sites are assumed to be closed to changes in occupancy (i.e., ‘closure’ implies no local extinction or colonization; MacKenzie et al. 2002). If the species is detected during a survey within a period where closure is assumed, the site is considered occupied. Hence, surveys at occupied sites that did not result in positive detection within a closed period can only be caused by imperfect detection (i.e., false-absences, false-zeros). Failure to detect the target species at a site within a closed period could result from true species absence or the species was present, but undetected during all sampling events. Investigators must use knowledge of the species biology to inform the study design with respect to the closure assumption (MacKenzie et al. 2006:104-106). In this regard occupancy studies share a similar sampling framework with mark-recapture research, which require repeated samples to estimate detectability of individual organisms within a closed period (Williams et al. 2002). Unlike mark-recapture research, however, relatively little is known about how violations of the ‘closure assumption’ can bias occupancy or detection estimates (but see Rota et al. 2009). Here I evaluate how the physical act of repeated sampling and seasonal changes in occupancy status result in bias if closure is assumed within a field season. The physical act of sampling may bias occupancy or detection estimates if destructive techniques are used or if sampling alters the behavior of the target species during subsequent 66      sampling events. Altered organism behavior as a consequence of sampling is typically referred to as a ‘trap-response’ in mark-recapture literature and multiple models have been developed to account for its potential bias (Zippin 1956, Otis et al. 1978). However, bias resulting from a ‘trap-shy’ response has not been tested in occupancy research. Furthermore, it is unknown to what degree invasive sampling techniques, which are often utilized for sampling cryptic species such as amphibians, reptiles and birds, can bias occupancy and detection estimates if not properly accounted for (Marsh and Goicochea 2003, Pike et al. 2010, Manning and Kaler 2011). Here, sampling can be viewed as an anthropogenic disturbance where the organism’s habitat or behavior is altered as a byproduct of investigators gathering species detection information. Examples of commonly used invasive sampling techniques include natural (i.e., rocks, logs, leaflitter) or artificial (i.e., wooden boards, sheet metal) cover object surveys for herpetofauna (reviewed in Dodd 2010) and intensive monitoring at avian nest sites or territories (Bolduc and Guillemette 2003, Manning and Kaler 2011). Occupancy studies are often conducted over time periods (e.g., weeks, months) where it may be inappropriate to assume closure for some species. For example, sampling often occurs during the breeding season for many species (e.g., pond-breeding amphibians, migratory birds, spawning fish). If the species synchronously arrives to all sites and sampling events correspond to this time period, then the closure assumption is likely met. However if organism arrival or departure is asynchronous, or if sampling events do not correspond to the time period when occupancy is static, the resulting non-random process of species availability may cause bias in occupancy estimates obtained via single-season occupancy models (MacKenzie et al. 2002, MacKenzie et al. 2006). Thus, identifying the appropriate time-scale for assuming closure requires investigators to incorporate knowledge of species phenology and movement ecology 67      into the sampling design (MacKenzie et al. 2006). Unfortunately, planning for field studies is often based on convenient calendar dates (e.g., single month, field season, or year) rather than the ecology of the target organism (Rota et al. 2009). Past research has highlighted the importance of accounting for violations of closure in mark-recapture research (e.g., Schwarz and Stobo, 1997, Kendall 1999, Kendall and Bjorkland 2001); however, I am aware of only one study that investigated occupancy bias resulting from violations of closure within a single field season. Rota et al. (2009) found substantial bias in songbird occupancy estimates when failing to account for violations of closure over sampling periods that are typical of ornithological research (i.e., spring and summer within a single year). I used a sampling design that allowed me to evaluate potential changes in site occupancy for a terrestrial salamander during a time period what would normally be considered a single sampling season (i.e., data analyzed using a single-season model; MacKenzie et al. 2002). Specifically, I determined if repeated sampling events caused a chronic decrease in red-backed salamander (Plethodon cinereus) occupancy during the sampling season, or if sampling only caused a temporary reduction in availability of salamanders that persisted for a short time following a sampling event. I also investigated whether species occurrence changed within the season due to natural processes, like seasonal movement, and its influence on occupancy estimates obtained from single-season models. I generalized my findings by conducting a simulation study to evaluate the sensitivity of occupancy estimates to general violations of closure when failing to account for local extinction occurring within a season. I use my results to develop design- and model-based recommendations for minimizing bias caused by animal movements or sampling disturbances. 3.2 Methods 68      3.2.1 Study Area I conducted my study in the Cadillac-Traverse City area in the northwestern Lower Peninsula of Michigan, USA, in 2009. This area is characterized as a glacial outwash plain with porous, sandy soils (Albert 1995). Temperate coniferous forests dominated this landscape prior to European settlement, but have since been replaced by hardwoods. My study occurred on stateowned forest lands that are currently managed for aspen (Populus sp.) production and have a history of timber harvest. I sampled three forest stands that were 5-8 years post-harvest and 1528 ha in size. Dominate understory woody plants were aspen, red maple (Acer rubrum), and blackberry (Rubus spp.). 3.2.2. Site Selection I used a two-step approach to select sample sites (i.e., transects). First, I used HAWTH’S TOOLS (Hawth’s Tools, version 3.27, http://www.spatialecology.com/htools/, accessed 1 Feb 2009) in a Geographic Information System (ArcGIS 9.1; Environmental Systems Research Institute, Redlands, CA) to overlay each aspen stand with a sampling lattice comprised of 60 x 60 m cells (Figure 3.1). I minimized potential heterogeneity in salamander occupancy from local landscape factors by eliminating all lattice cells whose borders intersected or encompassed an unharvested forest edge, active logging road, off-road recreational vehicle trail, or wetland appearing on 2005 National Agricultural Imagery Program imagery (Michigan Department of Information Technology 2007). I considered all remaining lattice cells as potential sampling locations and randomly selected seven within three aspen stands (i.e., 21 total cells). All selected cells were spaced ≥60 m apart. I assumed that habitat conditions known to affect red-backed salamander occurrence were relatively constant within each 60 x 60m cell (e.g., amount and type of forest, elevation) with the exception of the number of natural cover objects (NCO) along each 69      transect. I felt this assumption was reasonable because each forest stand was clearcut 5-7 years ago, possessed similar vegetation characteristics, and had little topographic relief. Second, I systematically placed six, 20 x 2m natural cover object (NCO) transects within each 60 x 60m cell (Figure 3.1). Each transect was assigned to one of three transect “groups”, which dictated when salamander sampling would be initiated: early May (Group 1), mid-June (Group 2), or late July (Group 3). Terrestrial salamanders, almost exclusively red-backed salamanders, are active and available during this period (Petranka 1998, Otto and Roloff 2011), and thus, investigators would normally assume closure. Group 1 transects were oriented east to west, at the north and south ends of each cell. Group 2 transects were oriented north to south, at the east and west ends of each 60 x 60 m cell. I randomly selected, without replacement, which cardinal direction the two transects in Group 3 faced. I positioned all transects in the third group 5 m from, and running parallel to, transects Groups 1 or 2 (Figure 3.1). This design provided three groups of 42 independent transects (n = 126 transects). Grouping transects in this fashion allowed me to determine if salamander occupancy was influenced by repeated sampling disturbances or if occupancy varied across time within a single field season due to organism nonrandom movements. Each transect represented an independent site in my analyses. I sampled all transects within an aspen stand twice in a 24-hr period (i.e., two surveys), allowing me to use a dynamic occupancy model to formally test the closure assumption over a time period that closure would normally be assumed. Under this approach, the two surveys within 24-hr constituted a “single-season” (hereafter “primary period”: MacKenzie et al. 2003). For each survey, one observer searched for red-backed salamanders under cover objects >4 cm diameter, >15 cm long, and within 1 m of the transect centerline. I ensured that the same observer did not survey the same transect twice in a single primary period. Observers recorded 70      the number of cover objects they searched and released all salamanders at the point of capture. Observers replaced all intact cover objects to their original point of origin and reconstructed fragmented cover objects to the best of their ability. After conducting two surveys within 24 hrs., observers revisited each transect 7-10 days later and continued sampling in this fashion until transect Groups 1 and 2 were surveyed 14 and 12 times, respectively (i.e., 7 and 6 primary periods). I surveyed Group 3 transects, four times (i.e., 2 primary periods). My sampling and handling protocols were approved by the Michigan State University Animal Care and Use Committee (Animal Use Form no. 07/08-118-00). 3.2.3. Model Development and Analysis I generated detection histories of red-backed salamanders for each transect. I coded detection histories so that each primary period represented the same calendar days for all transect groups. For example, detection histories for transects in each of the respective groups could include: Group1-Transect1 10 11 10 00 00 00 00 -- --, Group1-Transect2 11 10 00 11 10 01 00 -- --, Group2-Transect3 -- -- -- 11 01 00 00 00 01, Group2-Transect4 -- -- -- 00 01 11 00 00 00, Group3-Transect5 -- -- -- -- -- -- -- 01 00, Group3-Transect6 -- -- -- -- -- -- -- 11 10, where “1” represents a positive detection of at least one red-backed salamander during a single survey, “0” represents non-detection, and “-“ represents a missing value, indicating that the transect was not surveyed. At Group1-Transect1, an observer detected ≥1 red-backed salamander during the first survey of the first primary period, but failed to detect a salamander during the second survey. During the second primary period (i.e., 7-10 days later), ≥1 71      salamanders were detected during both surveys. During the third primary period the species was detected during the first survey, but not the second. Observers did not detect salamanders during the fourth, fifth, sixth, or seventh primary periods and the transect was not sampled during the final two primary periods (i.e., not surveyed in July). Non-detections of salamanders during primary periods 4-7 could arise from (1) failure of the observer to detect red-backed salamanders when the transect (site) was occupied (i.e., false negative, false zero) or (2) localized extinction that resulted in the site becoming unoccupied, which would violate the closure assumption if it was applied to the entire sampling season. MacKenzie et al. (2003) developed 3 parameterizations for the dynamic occupancy model. I used the parameterization that allows direct estimation of time-specific occupancy and extinction probabilities (ψ(t), ε(t)) to test 3 hypotheses regarding possible changes in salamander occupancy and extinction probability over time. These hypotheses are common to any species that is seasonally available or may be influenced by the sampling process. First, I considered a hypothesis where occupancy (ψ) remained static over time, but varied as a function of the number of cover objects along a transect (ψ(Cover), and no site extinction (ε(=0)) occurred throughout the duration of my study (see below for a description of COVER). Support for this hypothesis would suggest the closure assumption was not violated thereby permitting use of a single-season model to fit data from the entire sampling season (MacKenzie et al. 2002); ψ(Cover),ε(=0)). I considered this my null hypothesis for comparison to dynamic occupancy hypotheses that included the processes causing changes in occupancy and extinction probabilities within the sampling season. I hypothesized that changes in occupancy within my sampling season could result from the displacement of individuals (cumulative sampling hypothesis) caused by the cumulative 72      effects of repeated sampling or from natural processes of salamanders moving underground during warmer, drier months (seasonal movement hypothesis). I represented my cumulative sampling hypothesis by including a DISTURBANCE parameter that reflected the total number of times a transect was surveyed prior to primary period i (e.g., for a model with a ψ(Disturbance), ε(Disturbance) structure, occupancy probability at time i is modeled as logit(ψi) = β0+ β1(Number of surveys prior to i), and extinction as logit(εi) = β2+ β3(Number of surveys prior to i)). A competing hypothesis is that the seasonal availability of salamanders under cover objects may naturally decrease over summer, leading to local extinction and decreasing occupancy estimates during later primary periods (seasonal movement hypothesis). Support for this hypothesis is based on observations of directional salamander movement from underneath cover objects to the leaf-litter and subterranean refugia to escape desiccation as summer progresses (Taub 1961, Heatwole 1962). To represent this hypothesis, I developed a TREND model structure to allow occupancy and extinction to vary linearly across time (e.g., for a model with a ψ(Trend), ε(Trend) structure, logit(ψi) = β0+ β1(i-1), logit(εi) = β2+ β3(i-1)). I hypothesized that DISTRURBANCE and TREND could operate independently, or as additive effects on both occupancy and extinction probabilities (e.g., ψ(Disturbance+Trend), ε(Disturbance+Trend)). In addition to my three primary hypotheses, I also believed that occupancy would be higher for transects with a greater number of cover objects because of the increased availability of refugia (McKenny et al. 2006). I modeled cover objects as a continuous covariate; however, these models had poor convergence, likely due to a heavily right-skewed distribution of cover object numbers. Therefore, I classified transects as having either high (>20 cover objects) or low 73      (≤20 cover objects) levels of cover and included COVER as a categorical covariate on occupancy and extinction (e.g., ψ(Cover), ε(Cover)). Ranking of my top-five models did not change when using COVER as a categorical versus continuous covariate. I used cover object tallies from my first four surveys to quantify COVER for each site because repeated disturbances resulted in fragmented cover objects, thereby inflating my counts. I explored whether salamanders were temporarily affected by NCO disturbances by fitting models where: (1) detection probabilities were set equal between the first and second surveys within a primary period (i.e., within the same 24 hours; p(.)) and (2) detection probabilities varied between the two surveys, anticipating lower detection probability during the st nd second survey (p(1 _2 )). For these models I assumed that detection probability was constant st nd among primary periods (i.e., p(.,.) and p(.,1 _2 ). Although my current study design does not allow me to infer a mechanism, it is reasonable to suspect that sampling temporarily disrupts the microhabitat beneath cover objects, resulting in salamanders avoiding these areas, or temporarily emigrating, during the second survey within each primary period (Kendall 1999). I considered this a site-effect that would occur regardless of whether red-backed salamanders were actually detected during the first survey, given they were present within the primary period. Incorporating the hypotheses described above I developed a candidate model set consisting of 17 models (Table 1). First, I fit three models consistent with the closure assumption st nd for the entire sampling season: 1) ψ(Cover), ε(=0), p(.,.), 2) ψ(Cover), ε(=0), p(.,1 _2 ) and 3) ψ(Cover), ε(=0), p(t). These models assume static occupancy for the entire sampling season, but allow variation among high and low cover transects. The first closed model assumed detection probability is constant throughout the sampling season, the second model allows detection 74      probability to vary between surveys conducted with 24 hours, the third model allows detection to vary for every survey occasion. Next, I fit models that relaxed the closure assumption and allowed occupancy and extinction probability to vary within my sampling season as a function of TREND, DISTURBANCE or COVER and the associated additive combinations of these covariates. I believed that if these covariates were influential in my salamander system that occupancy and extinction probability would both be affected. For each occupancy and extinction structure I fit models where detection probability was constant (p(.,.)) or varied between the first st nd and second surveys within a primary period (p(.,1 _2 )) . I analyzed my data using Program MARK (MARK, version 5.1, http://www.cnr.colostate.edu/~gwhite/mark/mark.htm, accessed 07 July 2010) and used Akaike’s Information Criterion, adjusted for small sample size (AICc), to rank models (Burnham and Anderson 2002). I used cumulative AIC weights (w+) and evaluation of 95% confidence intervals to determine relative importance of covariates and model parameters. I report model averaged estimates and unconditional 95% confidence intervals for all real parameters. Relative bias in occupancy estimates was calculated as ( − /  where estimated occupancy probability for my AIC ‘best’ model that assumes closure and is  is the estimated probability for my ‘best’ dynamic model during primary period 1 (Bailey et al. 2007). 3.2.4. Simulation Study To generalize my field study findings, I evaluated the sensitivity of single-season occupancy models to violations of the closure assumption by quantifying bias for scenarios where local extinction occurs within a primary period. Here, I envision a general sampling situation where occupancy of the species of interest declines throughout the sampling season, however, the 75      investigator wrongfully assumes the system is closed and uses a single-season occupancy model for data analysis. I used a dynamic occupancy framework to generate expected values and then fit the data using a single-season occupancy model. This approach allowed me to assess bias when multi-season data, which violate the closure assumption, were fit to single-season occupancy models (see Bailey et al. 2007 for more details and associated software). I considered cases with either high or low initial occupancy probabilities (= 0.7, 0.4) and moderate or low detection probabilities (p(.) = 0.4, 0.2). For each combination of  and p I allowed extinction probability (ε) to vary between 0.0 and 1.0 between each primary period, which consisted of two independent surveys. Although the focus of my simulation was on extinction, I also considered two colonization probabilities that reflect my study system and may be realistic for biological systems experiencing a chronic decrease in occupancy over time (γ = 0.0, 0.05). For all simulations I considered designs with two or four primary periods (S) and two surveys (x) within each primary period for N = 150 sites. Here, occupancy state is allowed to change via local extinction and colonization between primary periods only. These simulations mimic traditional sampling designs with a single observer for each site survey, similar to my salamander data, where the total number of independent surveys (T) is equal to S × x. I used a dynamic occupancy structure ( parameter values for . , γ(.),ε(.),p(.); MacKenzie et al. 2003), and known , γ , ε , p, S, and N , to generate expected values for all possible detection histories in Program GENPRES (Bailey et al. 2007, GENPRES, version 3.0, http://www.mbr-pwrc.usgs.gov/software/presence.html, accessed 29 March 2011). I analyzed these expected values data to approximate bias and precision via an analytic-numeric approach (Burnham et al. 1987) using a closed occupancy model where T was equal to x × S. I fit two single-season models to each set of expected values, one model where detection probability was 76      . , held constant  all surveys ( . , . and a second model that allowed detection probability to vary across ). The second model was included because MacKenzie et al. (2006, p. 104- 106) suggested that a time-dependent model may reduce bias caused by violations of closure. I calculated relative bias as (E( – /  where E( is the estimated occupancy probability from my closed models fit to each expected value data set and  is the ‘true’ initial occupancy value used to generate these data. 3.3. Results 3.3.1. Salamander Field Study The closure assumption over my sampling season was clearly not supported by the field data; the three models with constant occupancy and no extinction probability (i.e., closed models) received no weight and AIC values were ≥ 29 units greater than my best models. Occupancy bias for my best static model (ψ(Cover), ε(=0), p(t)) was -0.30 and -0.08 for low and high quantities of COVER, respectively, when compared to estimates for my ‘best’ dynamic model st nd (ψ(Disturbance+Trend+Cover), ε(Disturbance+Trend+Cover), p(.,1 _2 )). Model selection results also revealed strong support for a cumulative sampling effect that resulted in decreased occupancy estimates over a sampling season (i.e., DISTURBANCE effect, w+ = 0.99; Table 1 and 2). For example, among transects in the high COVER category, model averaged occupancy estimates in early June were ≈ 47% higher for transects in Group 2, which had never been previously surveyed, when compared to Group 1 transects (previous surveys = 6; Figure 3.2A). I also found evidence of cumulative sampling resulting in increased extinction probability; however, estimates were imprecise (Figure 3.3) 77      I also found strong support for models containing the TREND covariate (w+ = 0.87; Table 1 and 2) with occupancy estimates decreasing across primary periods for all transect groups (Figure 3.2). Among transects in the high COVER category, model averaged estimates of occupancy in May (Group 1) were ≈39% higher than those transects initiated in July (Group 3; Figure 3.2A). Extinction probability also increased between later primary periods; however, estimates were imprecise (Figure 3.3). Collectively, my results provide strong evidence of a chronic disturbance effect associated with cumulative NCO surveys and seasonal decreases in salamander occupancy. I also found evidence that cumulative sampling disturbances and sampling later in the field season resulted in increased site extinction probabilities, although associated estimates were imprecise. Model selection results revealed substantial support for a short-term disturbance effect: st models that allowed detection to vary between the 1 and 2 nd survey within a 24 hour period had cumulative model weight w+ = 0.94 (Table 1). As I predicted, detection probability decreased during the second survey within a primary period ( 1 = 0.33, 95% CI: 0.25 - 0.42; 2 = 0.22, 0.16 - 0.30; Table 2). Lastly, I found strong support for spatial variation in occupancy and extinction probabilities among transects related to COVER (Table 2): transects with >20 cover objects had higher occupancy estimates, but also higher extinction estimates, than those with ≤20 cover objects (Figure 3.2 and 3.3). 3.3.2. Simulation Study Single-season occupancy models generally showed sensitivity to violations of the closure assumption, with the strength and direction of bias varying between simulations (Figure 3.4). For example, when ε was high (e.g., 0.75), was typically positively biased for all closed 78      models regardless of colonization levels. The only exception was when γ = 0.0 and data were modeled with time-varying detection probabilities . , : here, bias was slightly negative (Figure 3.4E-H). For lower values of ε, bias was generally smaller but the direction of the bias was positive when some colonization also occurred, γ = 0.05, and was negative when γ = 0.0. As expected, from simulations with four primary periods had greater bias, but smaller standard errors, than those with only two primary periods (Figure 3.4; Appendix A.1). For most simulations the . , . , model yielded occupancy estimates that were less biased than the . model; however, there were several situations where the opposite was true. For example, when there were four primary periods, bias for than . , was generally greater for  . , . (Figure 3.4D,G,H). Bias was generally lower for moderate versus low values of ; a pattern that was consistent for all but one simulation (Figure 3.4C). The pattern of bias was nearly identical for both high and moderate levels of for moderate levels of when γ = 0, but bias was often higher when γ = 0.05. 3.4 Discussion I explored the potential for bias caused by two processes commonly linked to violation of the population closure assumption in occupancy studies: 1) destructive sampling techniques or 2) sampling occurring over a time period during which organism movements result in variable occupancy status. In these cases the closure assumption is violated, causing bias in occupancy estimates obtained from standard, single-season occupancy models. Consistent with these concerns, I found evidence of a chronic decrease in occupancy, increase in local extinction, and temporary decrease in detection probability, associated with repeated sampling disturbances in my salamander system. I also found evidence for changes in seasonal availability of salamanders that was independent of the sampling process. 79      Although destructive sampling techniques have been acknowledged as a potential source of sampling bias in occupancy research (Smith and Petranka 2000, Bailey et al. 2004, Otto and Roloff 2011), my research is the first to explicitly test for and quantify sampling disturbance effects. I emphasize this concept by acknowledging that the effects of sampling disturbances can be both short-term and temporary, or long-term and persistent or permanent. My analysis provides evidence that destructive sampling causes short- and long-term effects on salamander detection and occupancy probabilities, respectively. I observed a 31% decrease in detection probability during the second survey within a primary period, suggesting that salamanders temporarily migrated either horizontally (i.e., outside the transect width) or vertically (i.e., into the leaf-litter or soil) following the first NCO survey. These organisms likely returned or were available for capture during subsequent primary periods. Kendall (1999) demonstrated that abundance estimates generated using mark-recapture analyses remain unbiased when temporary emigration occurs, if the variation in detection is modeled and there are only two sampling occasions per primary period. If I apply Kendall’s (1999) results to dynamic occupancy models, then my estimates for occupancy should remain unbiased even if the species temporarily leaves the site following a sampling disturbance. Thus, by having only two surveys per primary period, and modeling heterogeneity in detection probability, I able to minimize occupancy bias. This finding is further supported by my simulation results, where little bias (<10%) occurs when the . , model is fit to scenarios with only two primary periods (S=2). Importantly, I found evidence of a long-term, persistent sampling effect resulting in a chronic decrease in salamander occupancy and an increase in local extinction probabilities among primary periods. My sampling efforts caused unavoidable fragmentation of woody cover objects and reduced their overall contact with the soil and leaf litter, an outcome which is typical 80      for NCO surveys (Dodd 2010). My salamander analysis showed that simply modeling variation in detection probability over time (e.g., including Julian data as a detection covariate) was not sufficient for reducing bias in occupancy estimates obtained from single-season models. Furthermore, biases in my simulation study were higher and more variable with more primary periods (S=4), even for the . , model. More broadly, a sampling disturbance can be viewed as any act that is initiated by an investigator during the sampling process that results in increased variation, or reduced precision, of a parameter of interest (e.g., occupancy or detection probability). For example, detection probabilities may increase following first detection in cases where baits or lures are used (e.g., carnivore studies, Thorn et al. 2009). A similar effect can occur in volunteer-based monitoring programs when observers anticipate species that they have previously seen or heard at sample sites (Riddle et al. 2010). Alternatively, intensive monitoring at sites could decrease occupancy over time if the repeated presence of an investigator alters organism behavior or increases predation risk (e.g., Bolduc and Guillemette 2003, Manning and Kaler 2011). All these processes create variation in occupancy or detection probabilities, that if not modeled appropriately lead to biased estimates of occupancy parameters. Although my analysis was limited to a single case-study, it highlights the importance of carefully considering model assumptions and sampling design, especially when investigators use invasive sampling techniques to gather species detection information. In addition to disturbance effects, I observed strong evidence of decreased occupancy across primary sampling periods in my salamander example, presumably due to changes in seasonal availability of salamanders underneath NCOs. A decrease in salamander occupancy across time is supported by past research showing that use of cover objects by red-backed 81      salamanders decreases with drying soil conditions during summer months (Taub 1961, Heatwole 1962). As summer progresses, a portion of red-backed salamander populations may move underground or into the leaf-litter and are therefore unavailable for sampling using late-season NCO surveys (Bailey et al. 2004, Dodd and Dorazio 2004). Movement of wildlife in response to environmental conditions or seasonal changes in behavior is expected, yet I am aware of only one study that investigated sensitivity of occupancy models to violation of closure when species make non-random movements over the duration of a field study (Rota et al. 2009). This is problematic considering a vast majority of occupancy studies take place over multiple months (i.e., a field season); a timescale where investigators may wrongfully assume closure and use a single-season model to analyze occupancy data. In their study of migratory songbirds, Rota et al (2009) determined that their study sites were likely open to changes in occupancy over timescales typical of other ornithological research. Similarly, my results highlight the importance of addressing closure in occupancy studies, even when sampling relatively sedentary organisms such as terrestrial salamanders. The effects of both sampling disturbance and seasonal availability in my study resulted in increased local extinction probability and decreased occupancy, with little or no colonization, over the sampling season. This is analogous to emigration-only movement described by Kendall (1999) for assessing bias in mark-recapture models. Consistent with Kendall’s (1999) findings, and predictions made by MacKenzie et al. (2006), my simulations revealed that failing to account for permanent species emigration within time periods where closure is wrongfully assumed can result in substantial bias in occupancy estimates. Still, my simulation also revealed situations where occupancy models are relatively robust to closure violations. This was particularly true when extinction probabilities and the number of primary periods were relatively 82      low. In many field studies local extinction probability may be relatively low within a single field season for non-migratory species in fairly stable habitats, where sampling methods are relatively noninvasive. The closure assumption in these situations may be appropriate. My salamander field study provides an example where local-extinction may be initially low, but dramatically increase later in the field season, and thus require both design- and model-based solutions for minimizing bias. Ultimately, I concur with suggestions made by MacKenzie et al. (2006) and Rota et al. (2009) that investigators should use phenological knowledge of their study species when establishing the length of a primary period and collect data in a manner that allows investigators to formally test for closure if violations are suspect (e.g., use multiple independent observers or successive short-term sampling as demonstrated here). Interestingly, my simulations provide some scenarios that refute the assertion of MacKenzie et al. (2006) that use of a time-dependent model,  . , , should reduce occupancy bias caused by violations of closure. Although occupancy estimates from the  were generally less biased than those from the  relative bias was higher for . , . , . , model . model, there were several cases where . Occupancy estimates from the  higher (and negative) bias relative to the . , . , model showed . model, for scenarios with moderate levels of extinction probability and no recolonization, especially in longer duration studies (4 primary periods). My values of N, , , and γ were chosen to reflect my biological system and are not exhaustive: it is therefore unclear as to how a time dependent model would perform across broader parameter space. My results suggest investigators should exercise caution when using a time dependent model to reduce bias caused by violations of closure and use simulations to investigate nuisances unique to their study systems. 83      I suggest investigators use design- and model-based approaches to address closure in species occupancy studies. By addressing closure a priori in the design phase, investigators will reduce their dependency on sophisticated modeling to reduce bias attributable to a suboptimal study design. My field and simulation studies suggest that investigators using destructive sampling techniques should minimize the number of repeated site surveys within a primary period (i.e., time period when assuming closure). For example, my simulations revealed that occupancy estimates were generally less biased for single-season study designs with four site surveys compared to designs with eight surveys. This pattern was consistent regardless of whether variation in detection probability was modeled (p(t)) or not (p(.)). Thus, my salamander field study likely represents an optimal design for minimizing bias associated with temporary sampling effects, as I had only two site surveys per primary period (i.e., season). I also suggest that investigators use phenological knowledge of the target species to develop sampling designs that minimize false absences associated with seasonal movement or availability. For example, when surveying red-backed salamanders, investigators should sample within a single month to minimize bias associated with non-random movements into the subterranean environment (Heatwole 1962, Otto and Roloff 2011), but allow sites >24 hr to recover from sampling disturbances. Limiting the number of repeated surveys to ≤4 within a 3-4 week primary period should minimize bias for my salamander system. However, if surveys must be conducted over a long time span, or if >4 destructive sampling events must be used, then potential changes in occupancy over time should be accounted for within a dynamic occupancy framework, as I have done here. Investigators may also consider using spatial sub-sampling to avoid temporal bias in occupancy studies; however, spatial sub-sampling introduces its own set of assumptions that must be met to avoid bias (Kendall and White 2009). Ultimately, investigators should consider a 84      combination of design- and model-based strategies for minimizing estimation bias and achieving strong inference in occupancy studies. 3.5 Acknowledgements I thank A. Coleman and R. Sterling for their assistance in the collection of field data. M. Donovan and K. Fitzpatrick helped develop project objectives and A. J. Kroll and an anonymous reviewer provided insightful comments on an earlier version. G. White and K. Burnham provided analysis suggestions. Support for this project was provided by the Michigan Department of Natural Resources-Wildlife Division. 85    Table 3.1. Selection results for models fit to detection of red-backed salamanders sampled using natural cover object (NCO) surveys in northern Michigan, 2009. ΔAICc represents the difference between AICc values for model i and the top-ranking model; w is the Akaike weight; K is the number of parameters; −2 l is twice the negative log-likelihood. ΔAICc w K −2 l ψ(Disturbance+Trend+Cover), ε(Disturbance+Trend+Cover), p(.,1 _2 ) 0.00 0.82 10 806.4 ψ(Disturbance+Cover), ε(Disturbance+Cover), p(.,1 _2 ) ψ(Disturbance+Trend+Cover), ε(Disturbance+Trend+Cover), p(.,.) ψ(Disturbance+Cover), ε(Disturbance+Cover), p(.,.) st nd ψ(Trend+Cover), ε(Trend+Cover), p(.,1 _2 ) 3.79 5.90 9.49 0.12 0.04 0.01 8 9 7 814.3 814.4 822.1 12.27 0.00 8 822.8 ψ(Trend+Disturbance), ε(Trend+Disturbance), p(.,1 _2 ) 17.83 0.00 8 828.4 ψ(Disturbance), ε(Disturbance), p(.,1 _2 ) ψ(Trend+Cover), ε(Trend+Cover), p(.,.) ψ(Disturbance), ε(Disturbance), p(.,.) ψ(Disturbance+Trend), ε(Disturbance+Trend), p(.,.) st nd ψ(Trend), ε(Trend), p(.,1 _2 ) ψ(Cover), ε(=0), p(t) ψ(Trend), ε(Trend), p(.,.) st nd ψ(Cover), ε(Cover), p(.,1 _2 ) ψ(Cover), ε(Cover), p(.,.) st nd ψ(Cover), ε(=0), p(.,1 _2 ) ψ(Cover), ε(=0), p(.,.) 18.05 18.27 23.74 23.78 0.00 0.00 0.00 0.00 6 7 5 7 832.7 830.9 840.4 836.4 25.09 29.04 31.08 0.00 0.00 0.00 6 6 5 839.7 822.9 847.8 31.58 37.35 0.00 0.00 6 5 846.2 854.0 39.21 44.18 0.00 0.00 4 3 857.9 864.9 Model st st st st nd nd nd 86    nd Table 3.2. Parameter estimates for red-backed salamanders sampled using natural cover object (NCO) surveys in northern Michigan, 2009.  βDisturbance , βTrend , and βCover are the beta estimates representing logit-linear effects of DISTURBANCE, TREND, and nd COVER on occupancy and extinction probabilities, respectively. β2nd_Survey is the beta estimate for the 2 survey within a primary period. I report estimates and 95% confidence intervals for each of the top three models (w >0.01). Occupancy βDisturbance Model st nd ψ(Disturbance+Trend+Cover), ε(Disturbance+Trend+Cover), p(.,1 _2 ) Lower 95% Confidence Interval Upper 95% Confidence Interval st nd -0.17 -0.26 -0.08 ψ(Disturbance+Cover), ε(Disturbance+Cover), p(.,1 _2 ) Lower 95% Confidence Interval Upper 95% Confidence Interval -0.17 -0.26 -0.08 -0.20 -0.36 -0.04 -0.24 -0.34 -0.14 ψ(Disturbance+Trend+Cover), ε(Disturbance+Trend+Cover), p(.,.) Lower 95% Confidence Interval Upper 95% Confidence Interval βTrend 87    βCover 0.98 0.11 1.86 1.28 0.35 2.21 -0.20 -0.36 -0.04 1.01 0.13 1.89   Table 2 (cont’d) Extinction β2nd_Suvey βDisturbance Model st nd ψ(Disturbance+Trend+Cover), ε(Disturbance+Trend+Cover), p(.,1 _2 ) Lower 95% Confidence Interval Upper 95% Confidence Interval st nd βTrend βCover 0.31 0.05 0.56 0.60 0.21 0.98 1.92 0.20 3.63 -0.55 -0.94 -0.16 1.25 -0.48 2.99 -0.54 -0.92 -0.15 ψ(Disturbance+Cover), ε(Disturbance+Cover), p(.,1 _2 ) Lower 95% Confidence Interval Upper 95% Confidence Interval 0.57 0.22 0.91 ψ(Disturbance+Trend+Cover), ε(Disturbance+Trend+Cover), p(.,.) Lower 95% Confidence Interval Upper 95% Confidence Interval 0.31 0.05 0.57 88    Detection 0.60 0.21 1.00 1.92 0.17 3.67 Figure 3.1. Sampling lattice of 60 × 60-m cells overlaid on an aspen stand and systematically located sampling transects (in one random cell) for estimating red-backed salamander occupancy, extinction, and detection probability in northern Michigan, USA, 2009. Dark gray polygons in the background image represent unharvested canopy trees following timber harvest. Solid, dashed, and dotted lines in the foreground image represent 20-m transects initially surveyed in May, June, and July, respectively. 89      Figure 3.2. Red-backed salamander occupancy probability for transects with (a) high or (b) low levels of natural cover objects (NCO≥ 20 or NCO <20, respectively). Black circles represent Group 1 transects initiated in early May and surveyed through early July; open squares are Group 2 transects initiated in mid June and surveyed through late July; gray triangles are Group 3 transects initiated in late July and surveyed through early August. Each primary period was separated by 7-10 days. All parameter estimates are model averaged. A Occupancy probability (95% CI) 1.0 0.9 0.8 0.7 0.6 0.5 May 0.4 June 0.3 July 0.2 0.1 0.0 0 1 1 2 2 3 3 4 4 5 5 6 6 7 7 Primary sampling period 90    8 8 9 9 10   Figure 3.2 (cont’d) B Occupancy probability (95% CI) 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 0 1 1 2 2 3 3 4 4 5 5 6 6 7 7 Primary sampling period 91    8 8 9 9 10   Figure 3.3. Site extinction probabilities for red-backed salamander transects with (a) high or (b) low levels of natural cover objects (NCO≥ 20 or NCO <20, respectively). Black circles represent Group 1 transects initiated in early May and surveyed through early July; open squares are Group 2 transects initiated in mid June and surveyed through late July; gray triangles are Group 3 transects initiated in late July and surveyed through early August. Each primary period was separated by 7-10 days. All parameter estimates are model averaged. A Extinction probability (95% CI) 1.0 0.9 0.8 0.7 0.6 May June July 0.5 0.4 0.3 0.2 0.1 0.0 0 1 1 2 2 3 3 4 4 5 5 6 6 Primary sampling period 92    7 7 8 8 9 9 10   Figure 3.3 (cont’d) B Extinction probability (95% CI) 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 0 1 1 2 2 3 3 4 4 5 5 6 6 7 7 Primary sampling period 93    8 8 9 9 10   Figure 3.4. Estimates of relative bias in closed, single-season occupancy models when sites are open to local extinction events between primary periods. = initial occupancy probability, S = the number of primary sampling periods (i) where the number of surveys, t = 1, 2..(2×S), γ = probability an unoccupied site at time i, becomes occupied at time i + 1, ε = probability an occupied site at time i, becomes unoccupied at time i + 1, p = probability of detecting the species during a single survey, given its presence at a site. Expected values were analyzed using timeindependent ( . , . ; solid line) and time-dependent ( models. Relative bias was calculated as bias = (E( estimated occupancy for the closed model and  ; dashed line) closed occupancy /  where E( is the is the ‘true’ initial occupancy value. 94    – . ,   Figure 3.4 (cont’d) Relative bias (occupancy) ψint = 0.4, S = 2, γ = 0.05 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 -0.1 -0.2 A p(.) = 0.2 p(.) = 0.4 p(t) = 0.2 p(t) = 0.4 0 0.2 0.4 0.6 0.8 Extinction probability 1   Relative bias (occupancy) ψint = 0.7, S = 2, γ = 0.05 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 -0.1 -0.2 B 0 0.2 0.4 0.6 0.8 Extinction probability 1 95      Figure 3.4 (cont’d) ψint = 0.7, S = 4, γ = 0.05 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 -0.1 -0.2 C 0 Relative bias (occupancy) Relative bias (occupancy) ψint = 0.4, S = 4, γ = 0.05 0.2 0.4 0.6 0.8 Extinction probability 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 -0.1 -0.2 1 0 E 0 0.2 0.4 0.6 0.8 Extinction probability 1 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 -0.1 -0.2 1 F 0 96    0.2 0.4 0.6 0.8 Extinction probability ψint = 0.7, S = 2, γ = 0.0 Relative bias (occupancy) Relative bias (occupancy) ψint = 0.4, S = 2, γ = 0.0 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 -0.1 -0.2 D 0.2 0.4 0.6 0.8 Extinction probability 1   Figure 3.4 (cont’d) 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 -0.1 -0.2 ψint = 0.7, S = 4, γ = 0.0 G 0 Relative bias (occupancy) Relative bias (occupancy) ψint = 0.4, S = 4, γ = 0.0 0.2 0.4 0.6 0.8 Extinction probability 1 H 0 97    0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 -0.1 -0.2 0.2 0.4 0.6 0.8 Extinction probability 1   APPENDIX 98      Appendix A.1. Standard errors for occupancy estimates in closed occupancy models (i.e., singleseason; MacKenzie et al. 2002) when sites are open to local extinction events between primary periods. See Figure 3.4 for description of symbols and simulation overview.   ψint = 0.4, S = 2, γ = 0.05 ψint = 0.7, S = 2, γ = 0.05 Standard Error (Occupancy) 0.4 0.4 0.35 0.35 0.3 0.3 0.25 0.25 0.2 0.2 0.15 0.15 0.1 0.1 0.05 0.05 0 0 0.2 0.4 0.6 0.8 1 Extinction Probability p(.)=0.2 p(.)=0.4 p(t)=0.2 p(t)=0.4 0 0 0.4 0.4 0.35 0.35 0.3 0.3 0.25 0.25 0.2 0.2 0.15 0.15 0.1 0.1 0.05 0.05 0 0 0.2 0.4 0.6 0.8 1 Extinction Probability 0 0 99    1 ψint = 0.7, S = 4, γ = 0.05 ψint = 0.4, S = 4, γ = 0.05 Standard Error (Occupancy) 0.2 0.4 0.6 0.8 Extinction Probability 0.2 0.4 0.6 0.8 Extinction Probability 1   Appendix A.1 (cont’d) ψint = 0.7, S = 2, γ = 0.0 ψint = 0.4, S = 2, γ = 0.0 0.4 0.35 0.35 0.3 0.3 0.25 0.25 0.2 0.2 0.15 0.15 0.1 0.1 0.05 Standard Error (Occupancy) 0.4 0.05 0 0 0 0.2 0.4 0.6 0.8 Extinction Probability 1 0 ψint = 0.4, S = 4, γ = 0.0 0.4 0.35 0.35 0.3 0.3 0.25 0.25 0.2 0.2 0.15 0.15 0.1 0.1 0.05 0.05 0 0 0 0.2 0.4 0.6 0.8 Extinction Probability 1 0 100    1 ψint = 0.7, S = 4, γ = 0.0 0.4 Standard Error (Occupancy) 0.2 0.4 0.6 0.8 Extinction Probability 0.2 0.4 0.6 0.8 Extinction Probability 1   LITERATURE CITED 101      LITERATURE CITED Albert, Dennis A. 1995. Regional landscape ecosystems of Michigan, Minnesota, and Wisconsin: a working map and classification. Gen. Tech. Rep. NC-178. St. Paul, MN: U.S. Department of Agriculture, Forest Service, North Central Forest Experiment Station. Jamestown, ND: Northern Prairie Wildlife Research Center Online. http://www.npwrc.usgs.gov/resource/habitat/rlandscp/index.htm (Version 03JUN1998). Bailey, L. L., Hines, J. E., Nichols, J. D., and D. I. MacKenzie. 2007. Sampling design trade-offs in occupancy studies with imperfect detection: Examples and software. Ecological Applications 17:281-290. Bailey, L. L., Simons, T. R., and K. H. Pollock. 2004. Estimating site occupancy and species detection probability parameters for terrestrial salamanders. Ecological Applications 14:692-702. Bolduc, F., and M. Guillemette. 2003. Human disturbance and nesting success of Common Eiders: interaction between visitors and gulls. Biological Conservation 110:77-83. Burnham, K. P., and D. R. Anderson. 2002. Model Selection and Multi-model Inference: A Practical Information-theoretic Approach. 2nd edn. Springer-Verlag, New York, New York, USA. Burnham, K. P., Anderson, D. R., White, G. C., Brownie, C., and K. P. Pollock. 1987. Design and analysis of methods for fish survival experiments based on release–recapture. American Fisheries Society Monograph 5:1-437. Collier, B. A., Morrison, M. L., Farrell, S. L., Campomizzi, A. J., Butcher, J. A., Hays, K. B., Mackenzie, D. I., and R. N. Wilkins. 2010. Monitoring Golden-Cheeked Warblers on Private Lands in Texas. Journal of Wildlife Management 74:140-147. Dodd, C.K. 2010. Amphibian Ecology and Conservation: A Handbook of Techniques. Oxford University Press, New York, USA. Dodd, C. K., and R. M. Dorazio. 2004. Using counts to simultaneously estimate abundance and detection probabilities in a salamander community. Herpetologica 60:468-478. Ferraz, G., Nichols, J. D., Hines, J. E., Stouffer, P. C., Bierregaard, R. O., and T. E. Lovejoy. 2007. A large-scale deforestation experiment: Effects of patch area and isolation on Amazon birds. Science 315:238-241. Geissler, P. H., and M. R. Fuller. 1987. Estimation of the proportion of area occupied by an animal species. Proceedings of the Section on Survey Research Methods of the American Statistical Association 1986:533-538. 102      Heatwole, H. 1962. Environmental Factors Influencing Local Distribution and Activity of the Salamander, Plethodon cinereus. Ecology 43:460-472. Kendall, W. L. 1999. Robustness of closed capture-recapture methods to violations of the closure assumption. Ecology 80:2517-2525. Kendall, W. L., and G. C. White. 2009. A cautionary note on substituting spatial subunits for repeated temporal sampling in studies of site occupancy. Journal of Applied Ecology 46:1182-1188. Kendall, W. L. and R. Bjorkland. 2001. Using open robust design models to estimate temporary emigration from capture-recapture data. Biometrics 57:1113-1122. Kroll, A. J., Risenhoover, K., McBride, T., Beach, E., Kernohan, B. J., Light, J., and J. Bach. 2008. Factors influencing stream occupancy and detection probability parameters of stream-associated amphibians in commercial forests of Oregon and Washington, USA. Forest Ecology and Management 255:3726-3735. MacKenzie, D.I., Nichols, J. D., Royle, J. A., Pollock K. H., Bailey, L. L., and J. E. Hines. 2006. Occupancy estimation and modeling: inferring patterns and dynamics of species occurrence. Academic Press, New York, New York, USA. MacKenzie, D. I., Nichols, J. D., Hines, J. E., Knutson, M. G., and A. B. Franklin. 2003. Estimating site occupancy, colonization, and local extinction when a species is detected imperfectly. Ecology 84:2200-2207. MacKenzie, D. I., Nichols, J. D., Lachman, G. B., Droege, S., Royle, J. A., and C. A. Langtimm. 2002. Estimating site occupancy rates when detection probabilities are less than one. Ecology 83:2248-2255. Manning, J. A., and R. S. A. Kaler. 2011. Effects of survey methods on Burrowing Owl behaviors. Journal of Wildlife Management 75:525-530. Marsh, D. M. and P. C. Trenham. 2008. Current trends in plant and animal population monitoring. Conservation Biology 22:647-655. Marsh, D. M., and M. A. Goicochea. 2003. Monitoring terrestrial salamanders: biases caused by intense sampling and choice of cover objects. Journal of Herpetology 37:460-466. Martin, J., Chamaille-Jammes, S., Nichols, J. D., Fritz, H., Hines, J. E., Fonnesbeck, C. J., MacKenzie, D. I., and L. L. Bailey. 2010. Simultaneous modeling of habitat suitability, occupancy, and relative abundance: African elephants in Zimbabwe. Ecological Applications 20:1173-1182. 103      Mattfeldt, S. D., Bailey, L. L., and E. H. C. Grant. 2009. Monitoring multiple species: Estimating state variables and exploring the efficacy of a monitoring program. Biological Conservation 142:720-737. Mazerolle, M. J., Desrochers, A., and L. Rochefort. 2005. Landscape characteristics influence pond occupancy by frogs after accounting for detectability. Ecological Applications 15:824-834. McKenny, H. C., Keeton, W. S., and T. M. Donovan. 2006. Effects of structural complexity enhancement on eastern red-backed salamander (Plethodon cinereus) populations in northern hardwood forests. Forest Ecology and Management 230:186-196. Michigan Department of Information Technology. 2007. National agricultural imagery program data (NAIP 2005). URL http://www.mcgi.state.mi.us/mgdl/ [Accessed 01 Feb 2009]. Nichols, J. D., and Karanth, K. U. 2002. Statistical concepts: assessing spatial distributions. Pages 29-38 in K. U. Karanth and J. D. Nichols, editors. Monitoring Tigers and Their Prey: A Manual for Wildlife Managers, Researchers, and Conservationists. Centre for Wildlife Studies, Bangalore, India. Nichols, J. D., Bailey, L. L., O'Connell, A. F., Talancy, N. W., Grant, E. H. C., Gilbert, A. T., Annand, E. M., Husband, T. P., and J. E. Hines. 2008. Multi-scale occupancy estimation and modelling using multiple detection methods. Journal of Applied Ecology 45:13211329. Otis, D. L., Burnham, K. P., White, G. C., and D. R. Anderson. 1978. Statistical inference of capture data on closed animal populations. Wildlife Monographs, Number 62. Otto, C. R. V., and G. J. Roloff. 2011. Using multiple methods to assess detection probabilities of forest-floor wildlife. Journal of Wildlife Management 75:423-431. Petranka, J. W. 1998. Salamanders of the United States and Canada. Smithsonian Institution Press, Washington, DC. Pike, D. A., Croak, B. M., Webb, J. K., and R. Shine. 2010. Subtle - but easily reversible anthropogenic disturbance seriously degrades habitat quality for rock-dwelling reptiles. Animal Conservation 13:411-418. Riddle, J. D., Mordecai, R. S., Pollock, K. H., and T. R. Simons. 2010. Effects of prior detections on estimates of detection probability, abundance, and occupancy. Auk 127:94-99. Rota, C. T., Fletcher, R. J., Dorazio, R. M., and M. G. Betts. 2009. Occupancy estimation and the closure assumption. Journal of Applied Ecology 46:1173-1181. Schwarz, C. J., and W. T. Stobo. 1997. Estimating temporary migration using the robust design. Biometrics 53:178-194. 104      Seamans, M. E., and R. J. Gutierrez. 2007. Habitat selection in a changing environment: The relationship between habitat alteration and spotted owl territory occupancy and breeding dispersal. Condor 109:566-576. Smith, C. K., and J. W. Petranka. 2000. Monitoring terrestrial salamanders: Repeatability and validity of area-constrained cover object searches. Journal of Herpetology 34:547-557. Taub, F. B. 1961. The distribution of red-backed salamanders, Plethodon c. cinereus, within the soil. Ecology 42:681-698. Thorn, M., Scott, D. M., Green, M., Bateman, P. W., and E. Z. Cameron. 2009. Estimating brown hyena occupancy using baited camera traps. South African Journal of Wildlife Research 39:1-10. Tyre, A. J., Tenhumberg, B., Field, S. A., Niejalke, D., Parris, K., and H. P. Possingham. 2003. Improving precision and reducing bias in biological surveys: Estimating false-negative error rates. Ecological Applications 13:1790-1801. Weller, T. J. 2008. Using occupancy estimation to assess the effectiveness of a regional multiplespecies conservation plan: Bats in the Pacific Northwest. Biological Conservation 141:2279-2289. Williams, B. K., Nichols, J. D., and M. J. Conroy. 2002. Analysis and Management of Animal Populations. Academic Press, San Diego, CA. Zippin, C. 1956. An Evaluation of the Removal Method of Estimating Animal Populations. Biometrics 12:163-189. Zuckerberg, B., Bonter, D. N., Hochachka, W. M., Koenig, W. D., DeGaetano, A. T., and J. L. Dickinson. 2011. Climatic constraints on wintering bird distributions are modified by urbanization and weather. Journal of Animal Ecology 80:403-413. 105      CHAPTER 4 SONGBIRD RESPONSE TO GREEN-TREE RETENTION PRESCRIPTIONS IN CLEARCUT FORESTS Abstract Green-tree retention prescriptions have been proposed as a silvicultural technique for reducing the negative impacts of clearcutting on mature forest-associated species, while still promoting habitat for early-successional species. I evaluated the effectiveness of green-tree retention for conserving mature forest-associated, generalist, and early-successional songbirds in aspen stands that were 1 to 15 years post-harvest. I compared species occupancy estimates at sites where green-tree retention levels were exceeded (>10% canopy cover), met (3-10%), or below (<3%) recommended retention guidelines. I also compared the relative importance of these levels of green-tree retention to forest stand age in explaining songbird occurrence. With the exception of Red-eyed Vireos, my results suggest that current green-tree retention guidelines are ineffective at increasing site occupancy of mature forest-associated songbirds in aspen harvest units. Additionally, site occupancy of early-successional and generalist songbirds were unaffected by these levels of retention. Rather, site occupancy probabilities for most forest birds showed a clear association with age since harvest (1-15 years); however, the strength and direction of the association varied by species. My research highlights the importance of considering forest stand age when assessing the impact of silvicultural prescriptions on forest songbirds. Future management objectives in aspen forests should reflect the value of clearcutting to earlysuccessional and generalist bird species, while recognizing that green-tree retention as measured in my study does little to promote occupancy of the mature forest-associated bird community. 106      Keywords: aspen, bird occupancy, clearcut, structural retention, timber harvest, silviculture Status: Otto, C.R.V., Roloff, G.J. (in-press). Songbird response to green-tree retention prescriptions in clearcut forests. Forest Ecology and Management 4.1. Introduction Conservation biologists recognize habitat loss as a leading threat to worldwide biodiversity (Schipper et al., 2008; Kraus et al., 2010). One mechanism for habitat loss that occurs in many forests worldwide is timber harvest, specifically the silvicultural prescription known as clearcutting. This is an even-aged management prescription that results in complete removal of harvestable trees and has been implicated in the declines of numerous mature forestassociated species (Thompson et al., 2003; Griesser et al, 2007; Semlitsch et al., 2009). However, clearcutting can also serve as an effective management tool for promoting early successional habitat or increasing habitat heterogeneity (Thompson and DeGraaf, 2001; Franklin et al., 2002; DeGraaf and Yamasaki, 2003). Clearcutting may be particularly important for migratory songbird species that have experienced recent population declines because of earlysuccessional habitat loss (Litvaitis, 1993; Brawn et al., 2001; Hunter et al., 2001). Thus, clearcutting potentially creates a management paradox because it promotes habitat for some wildlife species, while reduces it for others. This is problematic considering one of the main objectives of recent forest management initiatives is to protect and conserve biodiversity within intensively-managed ecosystems (Carey and Curtis, 1996; Sustainable Forestry Initiative, 2010; Forest Stewardship Council, 2012). To alleviate the negative effects of clearcutting on forest wildlife, harvest prescriptions have been developed that include retention of canopy structure within the harvest units (i.e., 107      structural retention). These retention prescriptions require the preservation of “biological legacies”, such as green-trees and snags, within a harvested unit, in an effort to increase stand structural complexity and provide refugia for forest wildlife (Franklin et al., 1997; Schieck and Song, 2006; Rosenvald and Lõhmus, 2008). In some ecosystems retention prescriptions more closely mimic the result of natural disturbances, relative to clearcutting, and yet promote habitat for early-successional species (DeGraaf and Yamasaki, 2003; Rosenvald and Lõhmus, 2008; Lencinas et al., 2009). Green-tree retention prescription have been formally integrated into state, federal, and private management organizations through the adoption of forest practices regulations, voluntary best management practices, and forest stewardship certification programs (Rochelle, 2008; Elbakidze et al., 2011). Although many of these organizations share similar wildlife conservation goals, there is substantial inter-organization variability, regarding the proposed quantity and spatial arrangement of retained structure within harvest units. For example, the Northwest Forest Plan in the Pacific Northwest requires that ≥ 15% of the harvest unit area be retained on all federal forest lands (summarized in Aubry et al., 1999), while the Michigan Department of Natural Resources (MDNR) recommends 3-10% retention on state-owned lands (Bielecki et al., 2006). This variability likely stems from differences in management agency policies, objectives, and limited scientific information that directly relates to the management area of interest (Rosenvald and Lõhmus, 2008). Evaluating the effectiveness of retention guidelines can help address management questions such as ‘are the current levels of retention sufficient for conserving wildlife?’ and ‘do the benefits of structural retention incurred by mature forest-associated species outweigh potential negative effects on early-successional species?’ 108      I evaluated avian community response to structural retention guidelines implemented by MDNR in recently (1 to 15 years old) harvested aspen (Populus sp.) stands in Michigan. I compared species occupancy dynamics within a single breeding season at sites where green-tree retention levels were exceeded (>10%), met (3-10%), or below (<3%) the recommended guidelines. I used occupancy models (MacKenzie et al., 2006) to test the hypothesis that harvested sites with >10% green-tree retention exhibit higher occupancy by mature forestassociated songbirds within a breeding season. I also tested whether >10% green-tree retention reduced occupancy of early-successional and generalist songbird species. I then compared the relative importance of green-tree retention to forest age (1 to 15 years old) in explaining songbird occurrence across sites. Finally, my study design and modeling approach allowed me to indirectly assess how patterns in site settlement by migratory songbirds within a breeding season were related to structural retention and forest stand age. 4.2. Methods 4.2.1. Study Area I conducted my study in the Cadillac-Traverse City area in the northwestern Lower Peninsula of Michigan, USA, in 2010-11. This area is characterized as a glacial outwash-plain with porous sandy soils, intermixed with ice-contact zones and moraines (Albert 1995). My study occurred on state-owned forest lands that were managed for aspen production. Within harvested stands the MDNR implemented green-tree retention prescriptions (Bielecki et al., 2006), which called for retention of 3-10% of canopy trees and retention of snags and coarse woody debris (CWD) whenever possible. Green-tree retention prescriptions consisted of single leave-tree and aggregate retention. Retained canopy trees consisted of mixed oak (Quercus spp.), eastern white pine (Pinus strobus), red maple (Acer rubrum), black cherry (Prunus 109      serotina), American beech (Fagus grandifolia), and aspen, whereas dominate understory woody plants were aspen, red maple, blackberry (Rubus spp.), black cherry, downy serviceberry (Amelanchier arborea), witch-hazel (Hamamelis virginiana), American beech, and hophornbeam (Ostrya carpinifolia). Prior to European settlement, the study landscape was dominated by eastern white and jack pine (Pinus banksiana) and prone to fire disturbances (Barnes and Wagner, 2004). 4.2.2. Site Selection My initial group of candidate areas for study consisted of all state-owned aspen stands within a 4 county area that were >8 ha in size and between 1 to 15 years post-harvest. To select potential survey sites within a stand, I used HAWTH’S TOOLS (Hawth’s Tools, version 3.27, http://www.spatialecology.com/htools/, accessed 1 Feb 2009) in a Geographic Information System (GIS; ArcGIS 9.1; Environmental Systems Research Institute, Redlands, CA) to overlay each aspen stand with a 60 x 60 m lattice. I eliminated all lattice cells whose borders intersected or encompassed an unharvested forest edge, active logging road, off-road recreational vehicle trail, or wetland that appeared on 2005 National Agricultural Imagery Program imagery (NAIP; Michigan Department of Information Technology 2007). For the remaining lattice cells, I digitized canopy cover of all retained green-trees within the harvest units using NAIP imagery (1998-2010 depending on stand age). I assigned each lattice cell to one of nine ‘retention-age’ groups defined by the amount of green-tree retention and age since harvest. I considered three levels of green-tree retention based on the proportion of canopy cover within the 60 × 60 m cell, which was related to the MDNR retention guidelines (Bielecki et al., 2006; <3%, 3-10%, and >10% canopy cover) and three groups for age since harvest (1-4, 5-9, and 10-15 years postharvest). 110      Power approximations (Bailey et al., 2007) from a 2009 songbird pilot study revealed that my study design should include, at minimum, 153 sampling sites, split evenly between the nine ‘retention-age’ groups (i.e., 17 sites/group). I randomly selected 17 lattice cells for each ‘retention-age’ group and established a point-center within each of the selected cells. This pointcenter represented the spatial location for my avian point-count surveys (i.e., study sites). The point-center of all selected lattice cells were >250m apart and >75m from the harvest unit boundary. Once my initial site selection was complete (n=153 sites), I then randomly selected an additional 100 lattice cells. Here, all lattice cells 250m < x < 5km from a previously selected cell were included in the candidate pool. I eliminated 31 sites after initial field visits because the dominant cover-type was not aspen, the site resided within an ephemeral wetland, or it was inaccessible. My final sample size was 222 sites unequally distributed among retention-age groups. 4.2.3. Bird Sampling To reduce travel time, I sampled 120 sites, which were in close proximity to one another, in 2010, and 102 new sites in 2011. At each site, a single observer preformed a 9-min pointcount that was sub-divided into 3- 3min sub-counts. Point-counts were conducted between onehalf hour after sunrise and no later than three and one-half hours after sunrise on days with appropriate weather conditions (i.e., no rain, no strong winds). Detection was recorded for all songbirds seen or heard within a 50m sampling radius of the point-count center. I chose a 50m sampling distance to reduce the chance of sampling birds in adjacent forest stands (i.e., reduce edge effects). All observers were trained on estimating songbird detection distances before the study commenced. Each site was initially visited in late-May and then revisisted every 10-14 days. Sampling occurred in this fashion until all sites had been visited 4 times within a 2 month 111      time span during the breeding season (late-May to mid-July). I assigned each bird species to one of three habitat guilds (early-successional, generalist, interior forest) based on habitat preferences provided by Alsop (2001) and Poole (2005). I define Generalist bird species as those which are likely to occur in mature forests, open woodlands, early-successional forests, and scrub-shrub habitats, while interior forest species are likely to occur in mature forests with developed canopies. 4.2.4. Habitat Covariates In GIS, I used NAIP imagery to quantify the proportion of canopy cover within 50m of point-center for all study sites. Although I used 60x60 m cells as the dimensions of my site selection lattice, I felt it was more biologically meaningful to align the scale of my habitat data to that of my songbird data (Smith et al. 2008). Therefore, I used the proportion of canopy cover within 50m of point center to reassign each study site to one of my nine ‘retention-age’ groups (Figure 4.1). 4.2.5. Model Generation and Analysis I generated a detection history for each songbird species at each study site. Detection histories for a single species (e.g., ovenbird, Seiurus aurocapilla) at 2 study sites might appear as: Site 1 100 111 000 101 Site 2 010 000 110 000 where “1” represents a positive detection of ≥1 ovenbird during a 3-min survey and “0” represents non-detection. Each group of 3 numbers represents a 9-min point-count where sites were assumed closed to changes in occupancy (i.e., unoccupied sites did not become occupied, or occupied sites did not become unoccupied). The 9-min point count constituted a primary- 112      period within a dynamic occupancy model, whereas the 3 sub-counts were temporal sampling replicates used for estimating detection probability (MacKenzie et al. 2003). I hypothesized that site occupancy status of my focal bird species may change between each primary period, which were separated by 10-14 days. Changes in occupancy within a breeding season may result from migratory birds preferentially selecting higher quality habitat as the season progresses (Betts et al. 2008). Past research has shown that migratory songbird movement (site settlement and vacancy) rates can be indirectly estimated using dynamic occupancy models (Kendall and Nichols, 2004; Betts et al., 2008; Rota et al., 2009). My study design and modeling approach allowed us to indirectly estimate how site settlement probabilities varied as a function of greentree retention and years since harvest, for each species. I used an alternative parameterization of a dynamic occupancy model (MacKenzie et al., 2003) to estimate occupancy (ψ) within each primary period and settlement (γ: referred to as “colonization” in MacKenzie et al. 2003) probabilities between primary periods. Occupancy represents the probability that a site is occupied during a 9min point count. Settlement represents the probability that an unoccupied site during primary period i becomes occupied before primary period i + 1. Extinction probability [hereafter “vacancy”] is a derived parameter in this model. My results should be viewed as indirect estimates of species movement within a breeding season, as I did not determine movement from physically marked individuals (Kendall and Nichols, 2004; Betts et al. 2008). For the model development, I tested 5 hypotheses, represented by 10 models, regarding possible differences in songbird site occupancy and settlement. My null hypothesis was that occupancy for a given species remained static and no site settlement occurred throughout the duration of my study (ψ(.),γ(=0)). My second hypothesis was that no site settlement occurred 113      throughout my study (γ(=0)), but site occupancy varied as a function of green tree retention level (ψ(Retention),γ(=0)) or stand age (ψ(Age),γ(=0)). My third hypothesis was that occupancy remained static ψ(.), but settlement (γ(.)) occurred between primary periods and these parameters were not influenced by green-tree retention level or stand age ((ψ(.),γ(.)). Fourth, I hypothesized that occupancy and/or settlement were influenced by the amount of green-tree retention within 50m of the study site. To represent this hypothesis, I constructed 3 models (ψ(Retention),γ(Retention); ψ(Retention),γ(.); ψ(.),γ(Retention)) which allowed site occupancy and/or settlement probabilities to vary between my three retention groups: below (<3% residual canopy cover), at (3-10%), or exceeded (>10%). Strong support for models that included this structure would suggest that green-tree retention influences occupancy dynamics for a particular species. Lastly, I hypothesized that occupancy and/or settlement were influenced by the time since a stand was harvested (i.e., stand age: (ψ(Age),γ(Age); ψ(Age),γ(.); ψ(.),γ(Age)). Support for models that included this structure would suggest that stand age influences species occupancy dynamics. For my last 3 hypotheses I assumed that my collection of study sites were in a state of equilibrium with respect to occupancy- that is, the number of sites that were settled between primary periods was comparable to the number sites that were vacated (i.e., stationary Markov process; MacKenzie et al., 2006 p. 208-209). Thus, my reported occupancy estimates represent an average probability of occurrence among the four primary sampling periods. I held detection probability constant for all models (p(.)). I expected the direction and strength of the relationship between occupancy and settlement, and my two covariates, RETENTION and AGE, to be different for each bird species group. For interior forest species I expected strong support for models that included RETENTION, and occupancy and settlement estimates to be positively related to green-tree 114      retention level. For generalist species I expected substantial model selection uncertainty and imprecise occupancy and settlement estimates with respect to RETENTION and AGE. For early-successional species I expected strong support for models that included AGE; with occupancy and settlement estimates negatively related to AGE. I also expected a negative relationship between occupancy and RETENTION for all early-successional species. I analyzed my data using Program MARK (MARK, version 5.1, http://www.cnr.colostate.edu/~gwhite/mark/mark.htm, accessed 07 July 2010) and used Akaike’s Information Criterion, adjusted for small sample size (AICc), to rank models (Burnham and Anderson, 2002). I used cumulative AIC weights (w+) and evaluation of 95% confidence intervals to determine relative importance of covariates and model parameters. I report model averaged estimates and unconditional standard errors (SE) for all real parameters (Burnham and Anderson, 2002, p.149-205). Model averaging is a standard approach for conditioning inferences on a priori models in a candidate set (Burnham and Anderson 2002). No goodnessof-fit test for multi-season occupancy models is currently available. 4.3. Results The mean percentage of canopy cover for my three RETENTION groups was 0.6 ± 0.09 (mean ± 1SE: Group <3%), 6.2 ± 0.3 (Group 3-10%), and 26.0 ± 1.5 (Group >10%). Green tree retention levels were >30% and >50% at 31 and 8 study sites, respectively. Although, I detected 66 bird species during this study, sufficient detection information to model occurrence existed for 20 species. For these 20 species, detection probability estimates during a 3-min sub-count ranged from 0.31-0.72 (Table 4.1). For the above species, I detected 6.5 ± 0.19 (mean ± 1SE) per site. The proportion of sites where each species was known to occur often varied between each 9-min point-count (Table 4.2). 115      For interior forest species, cumulative weights (w+) for models that included RETENTION ranged from 0.00 (Ovenbirds) to 0.99 (red-eyed vireos, Vireo olivaceus; Table 4.3). Model-averaged occupancy estimates were similar among all green-tree retention levels for all interior forest species, except for red-eyed vireos, which showed a strong positive correlation with retention (Figure 4.2A). Occupancy estimates for red-eyed vireos were approximately 215 and 370% higher at sites that met (3-10%) or exceeded (>10%) the MDNR’s green-tree retention guidelines when compared to sites that were below (<3%) the guidelines (Figure 4.2A). Models for two interior forest species, hermit thrush (Catharus guttatus) and scarlet tanager (Piranga olivacea), had problems converging; therefore, parameter estimates for these species should be interpreted with caution (Figure 4.2A and 4.3A). Among generalist and early-successional species, support for models that included RETENTION was generally low (12 species, w+ <0.1; Table 4.3); however, there was moderate support for RENTENTION models for 5 species (0.1< w+ <0.7; Table 4.3). Among the species for which RETENTION was moderately supported, only Baltimore orioles (Icterus galbula) and rose-breasted grosbeaks (Pheucticus ludovicianus) exhibited occupancy estimates that varied between the three green-tree retention groups (Figure 4.2B,C). For example, model-averaged occupancy estimates for Baltimore orioles were approximately 44 and 36% higher at sites that met (3-10%) or exceeded (>10%) the retention guidelines when compared to sites that were below (<3%) the guidelines (Figure 4.2B). Estimates for rose-breasted grosbeaks were approximately 9 and 8% higher at sites that met or exceed the retention guidelines; however, there was substantial overlap in the 95% confidence intervals (Figure 4.2C). I did not observe evidence of green-tree retention negatively influencing occupancy for any of my study species 116      (Figure 4.2). Collectively, I observed little evidence that green-tree retention promotes occupancy of interior forest species, or inhibits occupancy of early-successional and generalist species, respectively. Cumulative weights (w+) for models containing stand AGE were >0.75 for 4 of 5 earlysuccessional, 7 of 10 generalist, and 1 of 5 interior forest species (Table 4.3). I observed little support for models that included AGE for Baltimore orioles, red-eyed vireos, and scarlet tanagers (w+ <0.10; Table 4.3). There was substantial inter-specific variation in occupancy estimates with relation to my three stand age classes (Figure 4.2). For example, occupancy estimates declined with increasing stand age for all early-successional species; however, the most dramatic decline occurred between the 5-9 year and 10-15-year age class of sites (Figure 4.2D). Occupancy estimates for golden-winged warblers (Vermivora chrysoptera) were similar for 1-4 year and 5-9 year sites, but decreased by approximately150% and 135% when compared to 10-15 year sites, respectively. Some generalist species, such as brown-headed cowbirds (Molothrus ater), exhibited occupancy estimates that declined with harvest age, and yet Nashville warblers (Oreothlypis ruficapilla), another generalist species, showed a positive response (Figure 4.2B,C). Collectively, my results provide evidence that songbird occupancy dynamics are often influenced by stand age within the first 15 years of harvest in aspen forests. I found strong evidence that occupancy status of my study sites changed between primary periods for all species: models where site settlement probability was fixed at 0.0 (i.e., occupancy was static across all primary periods) received no weight (Table 4.3). Furthermore, the probability that an unoccupied site became occupied during a later primary period (i.e., settlement) was strongly related to harvest age for many species, and related to green-tree retention for red-eyed vireos (Figure 4.3). For example, site settlement estimates for red-eyed 117      vireos were greater by 130 and 370% for sites that met or exceeded the retention guidelines when compared to sites that were below the guidelines (Figure 4.3A). Settlement for most earlysuccessional species showed a negative response to increasing stand age, while generalist and interior forest species showed mixed responses, respectively (Figure 4.3). 4.4. Discussion Green-tree retention prescriptions have been proposed as a silvicultural technique for reducing the negative impacts of clearcutting on interior forest species while still promoting habitat for early-successional species and maintaining a sustainable timber industry (Franklin et al., 1997; DeGraaf and Yamasaki, 2003; Rosenvald and Lõhmus, 2008). Although structural retention prescriptions are being implemented into forestry certification programs (e.g., Bielecki et al., 2006; Rochelle, 2008; Elbakidze et al., 2011), there is relatively little scientific information regarding wildlife response to these prescriptions. My results provide somewhat conflicting evidence that green-tree retention levels observed in this study are effective at increasing site occupancy of interior forest songbirds, at least in aspen clearcuts <15 years old. With the exception of red-eyed vireos, I did not observe strong evidence of any species exhibiting increased occupancy probabilities at sites that met or exceeded the MDNR’s green-tree retention guidelines when compared to sites with retention levels below the guidelines. Rather, site occupancy probabilities for most songbirds showed a clear association with age since harvest (115 years), with the strength and direction of the association varied by species. Occupancy estimates for Baltimore orioles and rose-breasted grosbeaks were higher at sites with 3-10% and >10% retention when compared to sites with <3% retention; however, the strength of the response was relatively weak. A meta-analysis of green-tree retention studies by Rosenvald and Lõhmus (2008) showed that the strength and direction of wildlife response to retention is often 118      conditional on retention quantity, cover-type, and the focal wildlife taxa, making comparisons across studies difficult. Similarly, my findings contradict other studies which documented a positive response of songbirds to green-tree retention, relative to clearcutting (e.g., Merrill et al., 1998; Atwell et al., 2008; Söderström, 2009), but support other studies that did not observe an effect (reviewed in Rosenvald and Lõhmus, 2008). Sites that exceeded the MDNR’s retention guidelines had an average of 25% canopy cover: a quantity of retention comparable to other studies that documented positive responses of songbirds to retention quantity (Atwell et al., 2008; Söderström, 2009). The landscape matrix of my study area consists of mixed northern hardwoods and conifer forests (Barnes and Wagner, 2004). Therefore the relative benefit of retention within my small (20-80 ha) harvest units may be reduced because forested habitat is a dominant feature in the landscape. It is unclear how the ecological role of structural retention may change in landscapes with different land-use conditions or within larger harvest units than those observed here. These relationships warrant further study. 4.4.1. Interior Forest Bird Response to Green-tree Retention Several interior forest species, such as Black-throated Blue Warblers (Setophaga caerulescens) and Northern Parula (Setophaga americana), were rarely detected at my study sites, even though they were present on the landscape. Lack of detection information precluded us from including these species in my analysis. Furthermore, occupancy estimates for three of my interior forest species were low, and the reliability of these estimates were suspect for Hermit Thrushes and Scarlet Tanagers. These findings suggest that interior forest species rarely occurred within my clearcuts, regardless of green-tree retention. Although inferred rarity of interior-forest species in clearcuts seems like an obvious finding, I note that a main objective of green-tree retention prescriptions is often to provide refugia for species that require mature forest 119      (i.e., life-boating; Franklin et al., 1997; Rosenvald and Lõhmus, 2008). Quantities of structural retention that I observed appear inadequate for promoting occupancy of many interior forest species in aspen clearcuts, except for Red-eyed Vireos. Twedt and Somershoe (2009) reported that densities of Red-eyed Vireos were lower in stands where variable retention prescriptions were applied, relative to untreated control stands. The objective of my study was to determine how occupancy of forest birds varied across three levels of green-tree retention that correspond to current MDNR guidelines, rather than comparing retention harvest to untreated “control” stands. Thus, my study shows that occupancy of Red-eyed Vireos is higher at sites where retention exceeded MDNR guidelines; however, it is unclear how occupancy at high retention sites compares to mature forest. A variety of environmental factors, interacting at multiple spatial and temporal scales, influence the occurrence of forest birds within a landscape (Sirami et al. 2008, Lesak et al. 2011, McDermott et al. 2011). Therefore identifying appropriate scales of forest management is of interest to natural resource managers. My analysis suggests that occupancy of most interior forest species is influenced by factors other than structural retention within harvest units. Consistent with this finding, McDermott et al. (2011) found that stand-level habitat factors, such as stand age and landform type, provided greater utility for predicting occurrence of latesuccessional birds in harvested forests when compared to within-stand vegetation factors. This suggests that management efforts for conserving interior forest bird should be focused at the stand and landscape scales, rather than the within-stand scale which was the focus of my study. My analysis did not include other within-stand habitat factors that could potentially influence occurrence of interior-forest birds such as green tree composition and size, and other habitat structures such as vertical snags (Poulin 2002, Remm et al., 2006, Jayapal et al. 2009). Canopy 120      green trees provide the only significant form of vertical structure within my harvest units: a feature that influences forest bird habitat selection and is of primary interest to forest management (Whelan, 2001; Poulsen, 2002; King et al., 2006). Thus, I feel my choice of green tree retention quantity as my primary habitat variable is justified, and acknowledge that other aspects of structural retention merit further study. 4.4.2. Competing Hypotheses: Songbird Response to Tree Retention, Relative to Stand Age I assessed evidence for multiple, competing hypotheses by evaluating songbird response to stand age, relative to varying green-tree retention levels, using an information theoretic (IT) approach. As with other studies that use IT, my inferences are conditional on the candidate model set (Burnham and Anderson, 2002). Had I not considered stand age (i.e., AGE models) in my analysis, then evidence for a positive influence of retention on occupancy of interior forest species would have been greater. This result highlights the importance of critical thinking and incorporating knowledge about the study system into development of a priori models. This was particularly true for all early-successional species, which showed a strong, negative response to years since harvest. Young aspen clearcuts with green-tree retention are used by a distinct avian community that require open or disturbed habitat for breeding (Merrill et al., 1998; DeGraaf and Yamasaki, 2003; Atwell et al., 2008). Clearcutting may serve as an effective management tool for creating early-successional habitat in landscapes where natural disturbances, such as fire, have been suppressed (Litvaitis, 1993; Brawn et al., 2001; Hunter et al., 2001). Collectively, my results corroborate those of Vanderwel et al. (2007), Campbell et al. (2007) and Donner et al. (2010) who stressed the importance of considering forest temporal dynamics when assessing the impact of silvicultural prescriptions on forest songbirds, even within the first 15 years following harvest. 121      4.4.3. Songbird Detection Probability and Intra-breeding Season Movement Failure to account for species detection has been cited as a leading source of uncertainty in wildlife-habitat research and monitoring programs (MacKenzie, 2005; Kéry and Schmidt, 2008; Watson et al. 2008). Even small amounts of observation error attributable to imperfect detection can bias results from aural point-count surveys if it is unaccounted for within an estimation framework (Etterson et al., 2009, McClintock et al., 2010). Consistent with this concern, I detected 66 species during this study, but had sufficient detection information to model occurrence patterns for just 20. Furthermore, detection probability for these 20 study species was always less than 1. Most studies investigating songbird response to timber harvesting use organism counts (incorrectly called abundance) or detection/non-detection (presence-absence or richness) data of multiple species as the response variable: data which are uncorrected for detection probability and thus possess an unknown level of bias (Kéry and Schmidt, 2008; McClintock et al., 2010). Future wildlife-forestry research should explicitly acknowledge this source of observation error and take steps to minimize potential bias in abundance, occupancy, or richness estimates by incorporating detection uncertainty into the modeling framework (Yoccoz et al. 2001, Royle et al., 2005; MacKenzie et al., 2006). I also observed evidence that the occupancy status of my study sites changed throughout the field season, presumably due to non-random movement of songbirds between primary periods. Although my analysis does not allow for direct inferences regarding species movements (Kendall and Nichols, 2004), it does suggest that habitat selection by my study species may occur as an iterative process, where sites with more suitable habitat conditions are more likely to be settled as the breeding season progressed. Migratory birds arriving on a landscape must make initial habitat selection choices based on limited knowledge of current conditions; however, these 122      choices may be refined as additional habitat information is gained (Hutto, 1985; Betts et al., 2008). Accordingly, my analysis revealed that site settlement probabilities within a breeding season were associated with stand age for many species and with green-tree retention for redeyed vireos. For example, site settlement probabilities for all of my early-successional species were lower on 10-15yr forest stands when compared to 1-4yr stands. Although telemetry and banding studies provide more direct evidence of species habitat selection, my study supports the growing use of patch occupancy models for inferring habitat selection dynamics of migratory birds in managed forests (Kendall and Nichols, 2004; Betts et al. 2008; Donner et al. 2010). Furthermore, my results corroborate those of Rota et al. (2009) who stress the importance of accounting for non-random movement of songbirds during point-count studies which span the length of a breeding season. 4.4.4. Conclusions My findings create a dilemma for resource managers using clearcutting with green-tree retention prescriptions for conserving songbirds in young aspen stands in Michigan. With the exception of Red-eyed Vireos, I found little evidence of green-tree retention influencing songbird occupancy when compared to the influence of forest succession within the first15 years of harvest. This pattern was consistent for all species guilds, including early-successional species, for which I hypothesized a negative association with green-tree retention. Based on these results, I do not support the notion that green-tree retention guidelines implemented by the MDNR are effective for promoting occupancy of most interior forest songbirds in harvested aspen stands. As with all scientific research, extrapolating these results to other systems should be done with caution. This is especially true for managed forests where green tree retention prescriptions may be greater than the quantities observed in my study (Aubry et al., 1999; 123      Hickey et al., 2001; Lance and Phinney, 2001; Martínez Pastur et al., 2009). For example, in some managed forests of Southern Patagonia, approximately 30% of the harvest area is retained as aggregate retention (Martínez Pastur et al., 2009). My study provides limited insight regarding the effectiveness green-tree retention within other forest systems where retention levels are higher than those observed here. However, retention levels observed in my study are comparable to other forest certification regulations in Finland, Norway, and Sweden and some portions of the Pacific Northwest (Arnott and Beese 1997; Aubry et al., 2004; Gustafsson et al. 2010). Because there is a high degree of inter-agency variability in management objectives related to, and application of, structural retention prescriptions, I feel it is prudent to provide management recommendations based on the observed evidence. If resource managers wish to promote occupancy of early-successional and generalist songbirds, then my results suggest that clearcutting, with limited structural retention, is a viable management option, particularly within the first 10 years of harvest. Although there is concern that canopy retention may reduce habitat quality for some early-successional birds (Tittler et al., 2001; McDermott and Wood, 2011), my study shows that <25% retention over a 0.78ha area did not result in lower occupancy of earlysuccessional species. If forest management objectives also include conservation of red-eyed vireos, then my results suggest that green-tree retention patches should provide ≥25% canopy cover within a 0.78ha area. However, it is unlikely that management goals for many interior forest songbirds will be accomplished solely through green-tree retention at the forest stand scale. Rather, successful management of these species may require conservation of entire stands of late-successional forests within actively managed landscapes. 4.5. Acknowledgements 124      I thank David Burt, Andrew Coleman, Jonathan Lautenbach, Joseph Lautenbach, Darin McNeil, Rachelle Sterling, Shayna Wieferich, and numerous volunteers for their assistance in the collection of field and GIS data. M. Donovan and K. Fitzpatrick (MDNR) helped develop project objectives. I thank M. Donovan and two anonymous reviewers for their thoughtful comments on an earlier version of this manuscript. Support for this project was provided by the MDNR Wildlife Division with funds from the federal Pittman-Robertson Wildlife Restoration Act grant administered by the United States Fish and Wildlife Service (W-147-R: Michigan’s Statewide Wildlife Research and Restoration Program). 125      APPENDIX 126      Appendix B.1 Methodology for calculating the proportion of canopy cover at study site. To estimate the proportion of canopy cover of each 50m radius site, I digitized canopy cover of all retained green-trees within the harvest units using 2010, 2009, 2005, and 1998 NAIP (National Agricultural Imagery Program) imagery. The age of the harvest unit dictated which year of NAIP imagery I used. For example, if harvest of a stand was completed in 2003, then I used 2005 NAIP imagery to map retention. To digitize canopy cover, I created a new shapefile in a Geographic Information System (GIS; ArcGIS 9.1; Environmental Systems Research Institute, Redlands, CA) to represent canopy cover provided by retained green-trees. I used the SKETCH tool in ArcEditor to draw the canopy cover polygons into the new shapefile. I then used the INTERSECT function in Arc Toolbox (ArcGIS 9.1; Environmental Systems Research Institute, Redlands, CA) to intersect the canopy cover polygon shapefile with separate polygon shapefile that represented the boundary of 50m radius site. I used the AREA function in HAWTH’S TOOLS (Hawth’s Tools, version 3.27, http://www.spatialecology.com/htools/) to calculate the proportion of canopy cover within each site. For example, if area of canopy cover 2 2 within a site was 1260m then the proportion of canopy cover at that site was {1260m / [ π × 2 (50m radius) ]} or 0.16. This was the value which I used to group my study sites for analysis in Chapter 4 (i.e., assign the site to one of three green-tree retention groups based on the MDNR retention guidelines (<3%, 3-10%, and >10% canopy cover)). 127      Figure 4.1. Number of songbird sampling sites, stratified by green-tree retention level and years since harvest. ‘Green-Tree Retention Level’ represents sites where the proportion of canopy cover within a 50m radius around the point-count station was below (<3%), at (3-10%), or exceeded (>10%) retention guidelines implemented by the MDNR in harvested aspen stands (Bielecki et al., 2006). ‘Years Since Harvest’ is the age of the site at the time sampling occurred. 128      Figure 4.2. Site occupancy estimates for 20 songbird species detected at 222 sites in young aspen stands from the northern Lower Peninsula of Michigan, USA, 2010-2011. Occupancy represents the probability that a site is occupied during a 9min point count. Green Tree Retention (%) is the percent canopy cover within a 50m radius of an avian point-count site; <3, 3-10, and >10% represent green tree retention levels that were below, met, or exceeded MDNR retention guidelines (Bielecki et al., 2006). Years Since Harvest is the age of the forest stand at the time of e sampling. All parameter estimates are model averaged. A = Interior Forest , B-C = f g Generalist , and D = Early-successional species. A Occupancy Probability ( 1SE) 0.45 0.4 0.35 Black & White Warbler 0.3 Hermit Thrush 0.25 0.2 Ovenbird 0.15 Red-eyed Vireo 0.1 Scarlet Tanager 0.05 0 0.5 Green-tree Retention (%) 1-4 5-9 10-15 Years Since Harvest     129        Figure 4.2 (cont’d) B Occupancy Probability ( 1SE) 0.45 0.4 0.35 American Redstart 0.3 0.25 American Robin 0.2 Baltimore Oriole 0.15 Black-capped Chickadee 0.1 Brown-headed Cowbird 0.05 0 0.5 Green-tree Retention (%) 1-4 5-9 10-15 Years Since Harvest   C Occupancy Probability ( 1SE) 0.45 0.4 0.35 Cedar Waxwing 0.3 Nashville Warbler 0.25 0.2 Rose-breasted Grosbeak 0.15 Veery 0.1 White-throated Sparrow 0.05 0 0.5 Green-tree Retention (%) 1-4 5-9 Years Since Harvest 10-15   130      Figure 4.2 (cont’d) D Occupancy Probability ( 1SE) 0.45 0.4 0.35 Chestnut-sided Warbler 0.3 Common Yellowthroat 0.25 0.2 Eastern Towhee 0.15 Golden-winged Warbler 0.1 Indigo Bunting 0.05 0 0.5 Green-tree Retention (%) 1-4 5-9 10-15 Years Since Harvest   e Interior forest birds; black-and-white warbler (Mniotilta varia), hermit thrush (Catharus guttatus), ovenbird (Seiurus aurocapillus), red-eyed vireo (Vireo olivaceus), scarlet tanager (Piranga olivacea) f Generalist birds: American redstart (Setophaga ruticilla), American robin (Turdus migratorius), Baltimore oriole (Icterus galbula), black-capped chickadee (Poecile atricapillus) brown-headed cowbird (Molothrus ater), cedar waxwing (Bombycilla cedrorum), Nashville warbler (Oreothlypis ruficapilla), rose-breasted grosbeak (Pheucticus ludovicianus), veery (Catharus fuscescens), white-throated sparrow (Zonotrichia albicollis) g Early-successional; chestnut-sided warbler (Setophaga pensylvanica), common yellowthroat (Geothlypis trichas), eastern towhee (Pipilo erythrophthalmus), golden-winged warbler (Vermivora chrysoptera), indigo bunting (Passerina cyanea) 131      Figure 4.3. Site settlement estimates for 20 songbird species detected at 222 sites in young aspen stands from the northern Lower Peninsula of Michigan, USA, 2010-2011. Settlement is the probability that an unoccupied site during primary period i becomes occupied before primary period i + 1. Green Tree Retention (%) is the percent canopy cover within a 50m radius of an avian point-count site; <3, 3-10, and >10% represent green tree retention levels that were below, met, or exceeded MDNR retention guidelines (Bielecki et al., 2006). Years Since Harvest is the age of the forest stand at the time of sampling. All parameter estimates are model averaged. A e f g = Interior Forest , B-C = Generalist , and D = Early-successional species. A Settlement Probability ( 1SE) 0.35 0.3 Black & White Warbler 0.25 0.2 Hermit Thrush 0.15 Ovenbird 0.1 Red-eyed Vireo 0.05 Scarlet Tanager 0 0.5 Green-tree Retention (%) 1-4 5-9 10-15 Years Since Harvest     132        Figure 4.3 (cont’d) B Settlement Probability ( 1SE) 0.35 0.3 American Redstart 0.25 0.2 American Robin 0.15 Baltimore Oriole 0.1 Black-capped Chickadee 0.05 Brown-headed Cowbird 0 0.5 Green-tree Retention (%) 1-4 5-9 10-15 Years Since Harvest   C Settlement Probability ( 1SE) 0.35 0.3 Cedar Waxwing 0.25 Nashville Warbler 0.2 Rose-breasted Grosbeak 0.15 Veery 0.1 White-throated Sparrow 0.05 0 0.5 Green-tree Retention (%) 1-4 5-9 Years Since Harvest 10-15   133      Figure 4.3. (cont’d) D Settlement Probability ( 1SE) 0.35 0.3 0.25 Chestnut-sided Warbler 0.2 Common Yellowthroat Eastern Towhee 0.15 Golden-winged Warbler 0.1 Indigo Bunting 0.05 0 0.5 Green-tree Retention (%) 1-4 5-9 10-15 Years Since Harvest     e Interior forest birds; black-and-white warbler (Mniotilta varia), hermit thrush (Catharus guttatus), ovenbird (Seiurus aurocapillus), red-eyed vireo (Vireo olivaceus), scarlet tanager (Piranga olivacea) f Generalist birds: American redstart (Setophaga ruticilla), American robin (Turdus migratorius), Baltimore oriole (Icterus galbula), black-capped chickadee (Poecile atricapillus) brown-headed cowbird (Molothrus ater), cedar waxwing (Bombycilla cedrorum), Nashville warbler (Oreothlypis ruficapilla), rose-breasted grosbeak (Pheucticus ludovicianus), veery (Catharus fuscescens), white-throated sparrow (Zonotrichia albicollis) g Early-successional; chestnut-sided warbler (Setophaga pensylvanica), common yellowthroat (Geothlypis trichas), eastern towhee (Pipilo erythrophthalmus), golden-winged warbler (Vermivora chrysoptera), indigo bunting (Passerina cyanea) 134      Table 4.1. Estimated detection probability and standard error estimates ( ̂ ± 1SE) for 20 songbird species detected in harvested aspen stands in the northern Lower Peninsula of Michigan, 2010-11. ̂ is the estimated probability of detecting the species during a 3 minute point-count, given it was present within a 50m radius of point-count site. ̂ 1SE Scientific Name Interior Forest Black-and-white warbler Hermit thrush Ovenbird Red-eyed vireo Scarlet Tanager 0.60 0.68 0.70 0.62 0.50 0.04 0.06 0.02 0.03 0.06 Mniotilta varia Catharus guttatus Seiurus aurocapillus Vireo olivaceus Piranga olivacea Generalist American redstart American robin Baltimore oriole Black-capped chickadee Brown-headed cowbird Cedar waxwing Nashville warbler Rose-breasted grosbeak Veery White-throated sparrow 0.72 0.43 0.51 0.56 0.31 0.37 0.65 0.61 0.60 0.52 0.02 0.04 0.05 0.03 0.04 0.04 0.02 0.02 0.03 0.05 Setophaga ruticilla Turdus migratorius Icterus galbula Poecile atricapillus Molothrus ater Bombycilla cedrorum Oreothlypis ruficapilla Pheucticus ludovicianus Catharus fuscescens Zonotrichia albicollis Species 135      Table 4.1 (cont’d) Species Early-successional Chestnut-sided warbler Common yellowthroat Eastern towhee Golden-winged warbler Indigo bunting ̂ 1SE Scientific Name 0.66 0.70 0.61 0.62 0.62 0.03 0.03 0.03 0.04 0.02 Setophaga pensylvanica Geothlypis trichas Pipilo erythrophthalmus Vermivora chrysoptera Passerina cyanea 136      Table 4.2. The proportion of 222 study sites where the target species was known to occur (i.e. positive detection) during 4, 9-minute point-counts. Each point-count was separated by 10-14 days. “Combined” represents the proportion of sites where the species was detected at least once during any of the 4 point-counts. Proportion of sites with known occurrence Species 1st 2nd 3rd 4th Combined Mature forest-associated Black-and-white warbler Hermit thrush Ovenbird Red-eyed vireo Scarlet tanager 0.11 0.02 0.21 0.05 0.03 0.09 0.02 0.15 0.14 0.03 0.08 0.03 0.19 0.11 0.03 0.05 0.03 0.21 0.17 0.05 0.23 0.08 0.42 0.34 0.12 Generalist American redstart American robin Baltimore oriole Black-capped chickadee Brown-headed cowbird Cedar waxwing Nashville warbler Rose-breasted grosbeak Veery White-throated sparrow 0.12 0.05 0.08 0.10 0.12 0.02 0.33 0.31 0.04 0.06 0.14 0.13 0.09 0.13 0.14 0.05 0.21 0.23 0.10 0.07 0.18 0.11 0.07 0.11 0.07 0.13 0.16 0.22 0.17 0.05 0.17 0.12 0.04 0.20 0.10 0.14 0.14 0.10 0.18 0.08 0.34 0.35 0.23 0.42 0.34 0.29 0.57 0.59 0.30 0.15 Early-successional Chestnut-sided warbler Common yellowthroat Eastern towhee Golden-winged warbler Indigo bunting 0.18 0.10 0.12 0.09 0.15 0.16 0.09 0.14 0.12 0.25 0.19 0.10 0.22 0.07 0.27 0.16 0.16 0.23 0.02 0.27 0.34 0.23 0.44 0.18 0.50 137      Table 4.3. Ranking of candidate occupancy models for 20 songbird species detected at 222 sites in young aspen stands from the northern Lower Peninsula of Michigan, USA, 2010-2011. ‘Retention’ is the percent canopy cover within a 50m radius of a avian point-count site, assigned to 1 of 3 classifications: sites where retention levels were below (<3%), at (3-10%), or exceeded (>10%) MDNR retention guidelines (Bielecki et al., 2006). ‘Age’ is the number of years post-harvest, assigned to 1 of 3 classifications: 1-4, 59, or 10-15 years post-harvest. I report the complete model set for Black-and-white Warblers, but report only models accounting for 0.80 of the total Akaike weight (wi) for all other species. Cumulative model weights (w+) for Retention (Reten) and Age are also reported. b Species Interior Forest Black-and-white warbler Hermit thrush Ovenbird Red-eyed vireo Model ψ(Age),γ(Age),p(.) ψ(Age),γ(.),p(.) ψ(.),γ(.),p(.) ψ(Retention),γ(.),p(.) ψ(.),γ(Retention),p(.) ψ(Retention),γ(Retention),p(.) ψ(.),γ(Age),p(.) ψ(Age),γ(=0),p(.) ψ(.),γ(=0),p(.) ψ(Retention),γ(=0),p(.) ψ(.),γ(.),p(.) ψ(Age),γ(Age),p(.) ψ(Retention),γ(Retention),p(.) Δ AICc 0.00 0.23 1.69 3.02 3.13 4.73 4.94 140.38 143.15 146.20 0.00 0.00 0.00 138    a w K −2 l w+ Reten w+ Age 0.34 0.30 0.15 0.07 0.07 0.03 0.03 0.00 0.00 0.00 0.85 1.00 1.00 7 5 3 5 5 7 5 4 2 4 3 7 7 733.8 738.1 743.6 740.9 741.0 738.5 742.8 880.3 887.1 886.1 287.7 1320.8 997.5 0.18 0.67 0.02 0.00 1.00 0.13 1.00 0.00   Table 4.2 (cont’d) b Species Scarlet tanager Generalist American redstart American robin Baltimore oriole Black-capped chickadee Brown-headed cowbird Cedar waxwing Nashville warbler Rose-breasted grosbeak Veery White-throated sparrow Model Δ AICc w K −2 l w+ Reten w+ Age ψ(.),γ(Retention),p(.) ψ(.),γ(.),p(.) 0.00 1.78 0.68 0.28 5 3 383.7 389.5 0.68 0.04 ψ(Age),γ(Age),p(.) ψ(.),γ(.),p(.) ψ(Age),γ(Age),p(.) ψ(.),γ(Age),p(.) ψ(.),γ(Retention),p(.) ψ(Retention),γ(.),p(.) ψ(Retention),γ(Retention),p(.) ψ(.),γ(.),p(.) ψ(.),γ(Retention),p(.) ψ(.),γ(.),p(.) ψ(.),γ(Age),p(.) ψ(Retention),γ(.),p(.) ψ(Age),γ(.),p(.) ψ(.),γ(Retention),p(.) ψ(Age),γ(Age),p(.) ψ(Age),γ(Age),p(.) ψ(Age),γ(Age),p(.) ψ(Age),γ(Age),p(.) ψ(Retention),γ(Retention),p(.) ψ(Age), γ(Age),p(.) ψ(.),γ(Age),p(.) ψ(.),γ(.),p(.) ψ(Age),γ(Age),p(.) 0.00 0.00 1.98 2.30 2.92 3.20 0.00 1.13 2.51 0.00 0.17 0.21 1.50 2.08 0.00 0.00 0.00 0.00 3.09 0.00 0.00 3.10 3.50 1.00 0.41 0.15 0.13 0.10 0.08 0.43 0.25 0.12 0.24 0.22 0.22 0.11 0.09 0.88 0.83 0.99 0.70 0.15 0.92 0.64 0.14 0.11 7 3 7 5 5 5 7 3 5 3 5 5 5 5 7 7 7 7 7 7 5 3 7 1136.8 953.5 947.4 951.7 952.4 952.6 674.7 683.9 681.2 1171.6 1167.7 1167.7 1169.0 1169.6 929.0 780.9 1567.5 1636.3 1639.4 1026.7 584.2 591.3 583.6 0.00 0.24 1.00 0.35 0.66 0.09 0.37 0.39 0.02 0.04 0.00 0.21 0.94 0.92 1.00 0.76 0.01 0.08 0.96 0.79 139    a   Table 4.2 (cont’d) b Species Model Early-successional Chestnut-sided warbler ψ(Age),γ(Age),p(.) Common yellowthroat Eastern towhee Golden-winged warbler Indigo bunting a K −2 l w+ Reten w+ Age 0.00 1.00 7 1247.7 0.00 1.00 ψ(.),γ(Retention),p(.) ψ(.),γ(Age),p(.) ψ(Age),γ(Age),p(.) 0.00 0.35 0.81 0.31 0.26 0.21 5 5 7 890.4 890.8 887.2 0.42 0.49 ψ(.),γ(.),p(.) ψ(Age),γ(Age),p(.) ψ(Age),γ(Age),p(.) ψ(Age),γ(Age),p(.) 2.43 0.00 0.00 0.00 0.09 1.00 0.86 1.00 3 7 7 7 896.9 1360.5 662.3 1637.1 0.00 0.02 0.00 1.00 0.95 1.00 ΔAICc = difference from the Akaike’s Information Criterion best model, adjusted for small sample size: K = no. of parameters; −2 l is twice the negative log-likelihood. b See Table 1 for scientific names. 140    a w Δ AICc   LITERATURE CITED 141      LITERATURE CITED Albert, Dennis A. 1995. Regional landscape ecosystems of Michigan, Minnesota, and Wisconsin: a working map and classification. Gen. Tech. Rep. NC-178. St. Paul, MN: U.S. Department of Agriculture, Forest Service, North Central Forest Experiment Station. Jamestown, ND: Northern Prairie Wildlife Research Center Online. http://www.npwrc.usgs.gov/resource/habitat/rlandscp/index.htm (Version 03JUN1998). Alsop III, F.J. 2001. Birds of North American: Eastern Region. D.K. Publishing, New York, USA. Arnott, J.T., Beese, W.J., 1997. Alternatives to clearcutting in BC Coastal Montane Forests. Forestry Chronicle 73, 670-678. Atwell, R.C., Schulte, L.A., Palik, B.J., 2008. Songbird response to experimental retention harvesting in red pine (Pinus resinosa) forests. Forest Ecol. Manage. 255, 3621-3631. Aubry, K.B., Amaranthus, M.P., Halpern, C.B., White, J.D., Woodard, B.L., Peterson, C.E., Lagoudakis, C.A., Horton, A.J., 1999. Evaluating the effects of varying levels and patterns of green-tree retention: Experimental design of the DEMO study. Northwest Sci. 73, 12-26. Bailey, L.L., Hines, J.E., Nichols, J.D., MacKenzie, D.I., 2007. Sampling design trade-offs in occupancy studies with imperfect detection: Examples and software. Ecol. Appl. 17, 281290. Barnes, B.V., Wagner Jr., W.H., 2004. Michigan Trees: A Guide to the Trees of the Great Lakes Region University of Michigan Press. Betts, M.G., Rodenhouse, N.L., Sillett, T.S., Doran, P.J., Holmes, R.T., 2008. Dynamic occupancy models reveal within-breeding season movement up a habitat quality gradient by a migratory songbird. Ecography 31, 592-600. Bielecki, J., J. Ferris, K. Kintigh, Koss, M., Kuhr, D., MacKinnon, S., Throop, S., Visser, L., Walters, M., 2006. Within stand retention guidance. Michigan Department of Natural Resources, Lansing, MI. http://www.michigan.gov/dnr/. Brawn, J.D., Robinson, S.K., Thompson, F.R., 2001. The role of disturbance in the ecology and conservation of birds. Annu. Rev. Ecol. Syst. 32, 251-276. Burnham, K. P., Anderson, D.R., 2002. Model Selection and Multi-model Inference: A Practical Information-theoretic Approach, second ed. Springer-Verlag, New York, USA. Campbell, S.P., Witham, J.W., Hunter, M.L., 2007. Long-term effects of group-selection timber harvesting on abundance of forest birds. Conserv. Biol. 21, 1218-1229. 142      Carey, A.B., Curtis, R.O., 1996. Conservation of Biodiversity: A Useful Paradigm for Forest Ecosystem Management. Wildl. Soc. B. 24, 610-620. DeGraaf, R.M., Yamasaki, M., 2003. Options for managing early-successional forest and shrubland bird habitats in the northeastern United States. Forest Ecol. Manage. 185, 179191. Donner, D.M., Ribic, C.A., Probst, J.R., 2010. Patch dynamics and the timing of colonizationabandonment events by male Kirtland's Warblers in an early succession habitat. Biol. Conserv. 143, 1159-1167. Etterson, M.A., Niemi, G.J., Danz, N.P., 2009. Estimating the effects of detection heterogeneity and overdispersion on trends estimated from avian point counts. Ecol. Appl. 19, 20492066. Elbakidze, M., Angelstam, P., Andersson, K., Nordberg, M., Pautov, Y., 2011. How does forest certification contribute to boreal biodiversity conservation? Standards and outcomes in Sweden and NW Russia. Forest Ecol. Manage. 262, 1983-1995. Forest Stewardship Council. 2012. FSC principles and criteria. On-line resource at http://www.fsc.org/pc.html. Accessed Jan 18, 2012. Franklin, J.F., Berg, D.R., Thornburgh, D.A., Tappeiner, J.C., 1997. Alternative silvicultural approaches to timber harvesting: Variable retention harvest systems, in: Kohm, K.A., Franklin, J.F. (Eds.), Creating a Forestry for the 21st Century: The Science of Ecosystem Management. Island Press, Washington D.C. pp. 111-138. Franklin, J.F., Spies, T.A., Van Pelt, R., Carey, A.B., Thornburgh, D.A., Berg, D.R., Lindenmayer, D.B., Harmon, M.E., Keeton, W.S., Shaw, D.C., Bible, K., Chen, J.Q., 2002. Disturbances and structural development of natural forest ecosystems with silvicultural implications, using Douglas-fir forests as an example. Forest Ecol. Manage. 155, 399-423. Griesser, M., Nystrand, M., Eggers, S., Ekman, J., 2007. Impact of forestry practices on fitness correlates and population productivity in an open-nesting bird species. Conserv. Biol. 21, 767-774. Gustafsson, L., Kouki, J., Sverdrup-Thygeson, A., 2010. Tree retention as a conservation measure in clear-cut forests of northern Europe: a review of ecological consequences. Scand. J. Forest Res. 25, 295-308. Hickey, J.E., Neyland, M.G, Bassett, O.D. 2001. Rationale and design for the Warra silvicultural systems trial in wet Eucalyptus obliqua forests in Tasmania. Tasforests 13, 155-182. Hunter, W.C., Buehler, D.A., Canterbury, R.A., Confer, J.L., Hamel, P.B., 2001. Conservation of disturbance-dependent birds in eastern North America. Wildl. Soc. B 29, 440-455. Hutto, R.L. 1985. Habitat selection by nonbreeding, migratory land birds, in: Cody, M.L. (Ed.), Habitat Selection in Birds. Academic Press, New York. Pp. 455-476. 143      Kendall, W.L., Nichols, J.D., 2004. On the estimation of dispersal and movement of birds. Condor 106, 720-731. Kéry, M., Schmidt, B.R., 2008. Imperfect detection and its consequences for monitoring for conservation. Comm. Ecol. 9, 207-216. King, D.I., Degraaf, R.M., Smith, M.L., Buonaccorsi, J.P., 2006. Habitat selection and habitatspecific survival of fledgling ovenbirds (Seiurus aurocapilla). J. Zool. 269, 414-421. Krauss, J., Bommarco, R., Guardiola, M., Heikkinen, R.K., Helm, A., Kuussaari, M., Lindborg, R., Ockinger, E., Partel, M., Pino, J., Poyry, J., Raatikainen, K.M., Sang, A., Stefanescu, C., Teder, T., Zobel, M., Steffan-Dewenter, I. 2010. Habitat fragmentation causes immediate and time-delayed biodiversity loss at different trophic levels. Ecol. Lett. 13, 597-605. Jayapal, R., Qureshi, Q., Chellam, R., 2009. Importance of forest structure versus floristics to composition of avian assemblages in tropical deciduous forests of Central Highlands, India. Forest Eco. Manage. 257, 2287-2295. Lance, A.N., Phinney, M., 2001. Bird responses to partial retention timber harvesting in central interior British Columbia. Forest Ecology and Management 142, 267-280. Lencinas, M.V., Martínez Pastur, G.J., Gallo, E., Cellini, J.M., 2009. Alternative silvicultural practices with variable retention improve bird conservation in managed South Patagonian forests. Forest Ecol. Manage. 258, 472-480. Lesak, A.A., Radeloff, V.C., Hawbaker, T.J., Pidgeon, A.M., Gobakken, T., Contrucci, K. 2011. Modeling forest songbird species richness using LiDAR-derived measures of forest structure. Remote Sens. Environ. 115, 2823-2835. Litvaitis, J.A., 1993. Response of early successional vertebrates to historic changes in land-use. Conserv. Biol. 7, 866-873. MacKenzie, D.I., Nichols, J.D., Hines, J.E., Knutson, M.G., Franklin, A.B., 2003. Estimating site occupancy, colonization, and local extinction when a species is detected imperfectly. Ecology 84, 2200-2207. MacKenzie, D.I., 2005. What are the issues with presence-absence data for wildlife managers? J. Wildl. Manage. 69, 849-860. MacKenzie, D.I., Nichols, J. D., Royle, J. A., Pollock K. H., Bailey, L. L., Hines, J. E., 2006. Occupancy Estimation and Modeling: Inferring Patterns and Dynamics of Species Occurrence. Academic Press, New York, USA. Martínez Pastur, G.J., Lencinas, M.V., Cellini, J.M., Peric, P.L., Soler Esteban, R., 2009. Timber management with variable retention in Nothofagus pumilio forests of Southern Patagonia. For. Ecol. Manage. 258, 436-443. 144      McClintock, B.T., Bailey, L.L., Pollock, K.H., Simons, T.R., 2010. Unmodeled observation error induces bias when inferring patterns and dynamics of species occurrence via aural detections. Ecology 91, 2446-2454. McDermott, M.E., Wood, P.B., 2011. Post-breeding bird responses to canopy tree retention, stand size, and edge in regenerating Appalachian hardwood stands. For. Ecol. Manage. 262, 547-554. McDermott, M.E., Wood, P.B., Miller, G.W., Simpson, B.T., 2011. Predicting breeding bird occurrence by stand- and microhabitat-scale features in even-aged stands in the Central Appalachians. For. Ecol. Manage. 261, 373-380. Merrill, S.B., Cuthbert, F.J., Oehlert, G., 1998. Residual patches and their contribution to forestbird diversity on northern Minnesota aspen clearcuts. Conserv. Biol. 12, 190-199. Michigan Department of Information Technology. 2007. National agricultural imagery program data (NAIP 2005). URL http://www.mcgi.state.mi.us/mgdl/ [Accessed 01 Feb 2009]. Poole, A. (Ed.). 2005. The Birds of North America Online: http://bna.birds.cornell.edu/BNA/. Cornell Laboratory of Ornithology, Ithaca, NY. Poulsen, B.O., 2002. Avian richness and abundance in temperate Danish forests: tree variables important to birds and their conservation. Biodivers. Conserv. 11, 1551-1566. Remm, J., Lohmus, A., Remm, K., 2006. Tree cavities in riverine forests: What determines their occurrence and use by hole-nesting passerines? For. Ecol. Manage. 221, 267-277. Rochelle, J.A., 2008. The biological basis for forest practices rules for retaining coarse woody debris in managed forests of the Pacific northwest. Tech. Rep. 954. National Council for Air and Stream Improvement, Olympia, WA, pp.11-16. Rosenvald, R., Lõhmus, A., 2008. For what, when, and where is green-tree retention better than clear-cutting? A review of the biodiversity aspects. Forest Ecol. Manage. 255, 1-15. Rota, C.T., Fletcher, R.J., Dorazio, R.M., Betts, M.G., 2009. Occupancy estimation and the closure assumption. J. of Appl. Ecol. 46, 1173-1181. Royle, J.A., Nichols, J.D., Kery, M., 2005. Modelling occurrence and abundance of species when detection is imperfect. Oikos 110, 353-359. Schieck, J., Song, S.J., 2006. Changes in bird communities throughout succession following fire and harvest in boreal forests of western North America: Literature review and metaanalyses. Can. J. Forest Res. 36, 1299-1318. Schipper, J., Chanson, J.S., Chiozza, F., et al., 2008. The status of the world's land and marine mammals: Diversity, threat, and knowledge. Science 322, 225-230. Semlitsch, R.D., Todd, B.D., Blomquist, S.M., Calhoun, A.J.K., Gibbons, J.W., Gibbs, J.P., Graeter, G.J., Harper, E.B., Hocking, D.J., Hunter, M.L., Patrick, D.A., Rittenhouse, 145      T.A.G., Rothermel, B.B., 2009. Effects of Timber Harvest on Amphibian Populations: Understanding Mechanisms from Forest Experiments. Bioscience 59, 853-862. Shiver, B.D., Borders, B.E., 1996. Sampling Techniques for Forest Resource Inventory. John Wiley and Sons Inc., USA. Sirami, C., Brotons, L., Martin, J.L., 2008. Spatial extent of bird species response to landscape changes: colonisation/extinction dynamics at the community-level in two contrasting habitats. Ecography 31, 509-518. Smith, K.M., Keeton, W.S., Donovan, T.M., Mitchell, B., 2008. Stand-level forest structure and avian habitat: Scale dependencies in predicting occurrence in a heterogeneous forest. Forest Sci. 54, 36-46. Söderström, B., 2009. Effects of different levels of green- and dead-tree retention on hemi-boreal forest bird communities in Sweden. Forest Ecol. Manage. 257, 215-222. Sustainable Forestry Initiative. 2010. Requirements for the SFI 2010-2014 program, Section 2: Sustainable Forestry Initiative 2010-2014 standard. Sustainable Forestry Initiative, Washington D.C. pp.14. Thompson, F.R., DeGraaf, R.M., 2001. Conservation approaches for woody, early successional communities in the eastern United States. Wildl. Soc. B. 29, 483-494. Thompson, I.D., Baker, J.A., Ter-Mikaelian, M., 2003. A review of the long-term effects of postharvest silviculture on vertebrate wildlife, and predictive models, with an emphasis on boreal forests in Ontario, Canada. Forest Ecol. Manage. 177, 441-469. Tittler, R., Hannon, S.J., Norton, M.R., 2001. Residual tree retention ameliorates short-term effects of clear-cutting on some boreal songbirds. Ecol. Appl. 11, 1656-1666. Twedt, D.J., Somershoe, S.G., 2009. Bird Response to Prescribed Silvicultural Treatments in Bottomland Hardwood Forests. J. Wildl. Manage. 73, 1140-1150. Vanderwel, M.C., Maicolm, J.R., Mills, S.C., 2007. A meta-analysis of bird responses to uniform partial harvesting across North America. Conserv. Biol. 21, 1230-1240. Watson, C.A., Weckerly, F.W., Hatfield, J.S., Farquhar, C.C., Williamson, P.S., 2008. Presencenonpresence surveys of golden-cheeked warblers: detection, occupancy and survey effort. Animal Conserv. 11, 484-492. Whelan, C.J., 2001. Foliage structure influences foraging of insectivorous forest birds: An experimental study. Ecology 82, 219-231. 146      CHAPTER 5 OCCURRENCE AND SURVIVAL OF A TERRESTRIAL SALAMANDER FOLLOWING HABITAT DEGRADATION: CAN WE AMELIORATE THE EFFECT OF TIMBER HARVEST? Abstract Habitat degradation resulting from anthropogenic activities poses immediate and prolonged threats to biodiversity, particularly among declining amphibians. Most studies infer amphibian response to habitat degradation by correlating patterns in species occupancy or abundance with an ecological effect, often with little regard to the demographic process underlying these patterns. I used correlative and experimental evidence to evaluate how retention of vertical green trees (CANOPY) and coarse woody debris (CWD) influenced terrestrial salamander occupancy and survival in clearcut forests. I observed little evidence that terrestrial salamander occupancy was related to structural retention within harvested forests: cumulative weight of models was 0.36 and 0.37 for CANOPY and CWD, respectively and model parameters consistently overlapped 0. However, salamander survival within recently harvested forests was positively influenced by CANOPY and CWD retention ( and = 0.73 (0.27 – 1.19; 95%CI) = 1.01 (0.53 – 1.50). In some instances, estimated salamander survival was higher at harvested sites with high levels of structural retention compared to older (40-60 year-old) forests. I demonstrated that organismal pattern (occupancy) and process (survival) can function independently in response to habitat change. My study emphasizes the importance of quantifying population vital rates when evaluating mitigation techniques presumed beneficial to wildlife. 147      Keywords: amphibian, capture probability, clearcutting, demography, coarse woody debris, detection, occupancy, survival, vital rates, timber harvest Status: I intend to submit this manuscript to Ecological Applications. 5.1. Introduction Anthropogenic habitat degradation is a primary threat to global biodiversity (Foley et al. 2005; Schipper et al. 2008; Krauss et al. 2010). For example, > 30% of amphibian species are at risk of extinction and habitat degradation is cited as a leading cause of amphibian declines (Stuart et al. 2004; Wake and Vredenburg 2008; Hof et al. 2011). In forested systems, degradation via timber harvesting poses both short- and long-term threats to forest biota (Ferraz et al. 2003; Gibson et al. 2011). Most researchers agree that timber harvesting has a negative effect on forest amphibian abundance (Petranka et al. 1993; Ash 1997; Knapp et al. 2003, Homyack and Haas 2009), thus amphibians are useful candidates for studying this form of habitat degradation on wildlife. Limited knowledge exists on how timber harvesting influences amphibian population vital rates because most observational and experimental studies use indices like species richness (counts of the number of species), relative abundance (counts of individuals within a species), or occurrence (counts of occupied sites) as the response variable (reviewed in deMaynadier and Hunter 1995; Cushman 2006). Use of these population indices for inferring response to habitat degradation has been criticized for several reasons (e.g., Cushman 2006, Welsh et al. 2008). First, assessing changes in abundance or occupancy does not provide adequate information on how population vital rates or life-history characteristics change as a consequence of habitat condition (Van Horne 1983; Yoccoz et al. 2001). When inferring species response to a particular habitat condition or ecological effect, researchers implicitly assume a positive, and 148      often linear, relationship exists between species occurrence, or abundance, and habitat (Van Horne 1983). Support for this assumption may be tenuous if amphibians exhibit metapopulation or source-sink dynamics (Carlson and Edenhamn 2000, Marsh and Trenham 2001, Heard et al. 2012). Furthermore, patterns in species occurrence and abundance are a dictated by population process such as survival, reproduction, and movement. Hence, studies that directly assess the influence of habitat degradation on survival, reproduction, and movement of organisms should yield greater inferential power that those which assess population indices (Yoccoz et al. 2001; Williams et al. 2002). Second, raw counts of individuals or species are often incorrectly referred to as (relative) abundance, or richness data; however, these data are not corrected for probability of encounter (i.e., detection probability; MacKenzie and Kendall 2002, Kéry and Schmidt 2008). Failure to account for detection probability has been cited as a leading source of uncertainty in research investigating amphibian response to timber harvest and other habitat relationships (Bailey eta l. 2004a, Mazerolle et al. 2005, Pellet and Schmidt 2005, Kroll 2009). These authors stress that count data that are uncorrected for detection probability possess an unknown level of uncertainty, and therefore should be viewed as indices of abundance or richness, which in turn, are indices of population processes. Over the past decade, a myriad of statistical techniques have been developed to estimate species occupancy and abundances from replicate count data (e.g., MacKenzie et al. 2002, Royle 2004, Royle and Dorazio 2004). These estimation techniques are generally preferred over traditional Generalized Linear Models (GLM) because they account for imperfect detection probability (Kéry and Schmidt 2008, Archaux et al. 2012). Thus, ecologists now possess the tools for generating unbiased estimates of occupancy and abundance from count 149      data which can be collected relatively quickly over broad spatial scales to address ecological and management questions. I used correlative and experimental approaches to investigate red-backed salamander (Plethodon cinereus; RBS) response to residual forest structure, such as coarse woody debris and vertical green trees, within harvested forests. These structures were purposefully left to potentially ameliorate the negative effects of clearcutting on forest wildlife (Franklin et al. 1997). First, I studied how patterns in salamander occupancy were related to retention of green trees (CANOPY) and coarse woody debris (CWD) within a managed, forested landscape (560,000ha). Second, I quantified how salamander survival was influenced by CANOPY and CWD within timber harvest areas. For both studies, I used hierarchical models that account for observation error associated with imperfect detection. By focusing my study on the same species, forested areas, and time period I was able to evaluate if occupancy and survival measurements yielded similar inferences regarding amphibian response to structural retention. Hence, my study informs the current debate over which population parameters are most useful for assessing amphibian response to habitat degradation (Cushman 2006; Welsh et al. 2008, Kroll et al. 2009). 5.2. Methods 5.2.1. Study Species Red-backed salamanders (RBS) are a terrestrial, lung-less amphibian distributed in woodlands throughout eastern North America (Petranka 1998, p. 336). RBS are recognized as critical components of forested ecosystems through their contribution to the detrital food web, forest biomass and potentiality serve as an indicator species (Davic and Welsh 2004). Like all Plethodontids, respiration in RBSs occurs cutaneously, making them prone to desiccation. As a result, RBS has been utilized in many forest management studies and its negative response to 150      clearcutting has been well documented (e.g., Petranka et al. 1993; Ash 1997; Homyack and Haas 2009). 5.2.2. Study Area I conducted my study across a 560,000 ha area in the northwestern Lower Peninsula of Michigan, USA, in 2010-11. My study occurred on state-owned forest lands that were managed for aspen (Populus spp.) production by the Michigan Department of Natural Resources (MDNR). In Michigan, aspen is typically harvested via clearcutting on a 50-70 year rotation. Within harvested stands (where a stand is defined as an area with homogenous vegetation and management focus) the MDNR implemented green-tree retention prescriptions to provide refugia for wildlife following timber harvesting (Franklin et al. 1997). These prescriptions called for retention of 3-10% of the pre-harvest green-tree basal area (i.e., the cumulative surface area covered by a cross-section of tree stems at ground level), arranged throughout the stand as single leave-trees or aggregated into retention patches (Bielecki et al. 2006). Harvested areas also contained varied amounts of CWD that was unequally distributed throughout the harvest areas as part of the logging contract specifications. Additional study area details can be found in Otto and Roloff (2011). 5.2.3. Salamander Occupancy My candidate areas for study consisted of all state-owned aspen stands within a 4 county area that were >8 ha in size and between 1 to 5 years post-harvest. To select potential survey sites within a stand, I used HAWTH’S TOOLS (Hawth’s Tools, version 3.27, http://www.spatialecology.com/htools/) in a Geographic Information System (GIS; ArcGIS 9.1; Environmental Systems Research Institute, Redlands, CA) to overlay each aspen stand with a 60 x 60 m lattice. I used orthophotos from the 2010 National Agricultural Imagery Program (NAIP; 151      http://www.mcgi.state.mi.us/mgdl) to digitize canopy cover of all retained green-trees within the sampling lattice over each forest stand. I assigned each cell of the lattice to one of three groups (>25%, 10-25%, and <10% canopy cover) and then randomly selected 40 cells from each group. To maintain site independence, I ensured that all selected cells were >200m apart. I selected an additional 30 cells within 40-60 year old aspen stands that were adjacent to my harvested stands. I eliminated 16 cells after initial field visits because the dominant cover-type was not aspen. My final sample size for harvested cells was 107, with varying levels of green-tree retention, and 27 for the 40-60 year-old cells. Hereafter, I refer to the subset of cells used for this study as sites. Within each selected site I identified 33, 20 x 2m transects which were oriented north to south and spaced ≥5m apart.  From the 33 transects I randomly selected, with replacement, 3 transects for salamander sampling. Selected transects were treated as spatial replicates for estimating salamander detection probability (MacKenzie et al. 2002; Kendall and White, 2009). Each transect was surveyed once unless it was selected with replacement, in which case it was surveyed again 12-16 days later. For each transect survey, one observer searched for RBSs under cover objects >4 cm diameter and >15 cm long. Observers tallied the number of cover objects they search along each transect. I generated detection histories of RBSs for each site and used single-season occupancy models (MacKenzie et al. 2002) to test predictions regarding salamander occupancy and detection. I started with a general hypothesis that salamander occupancy probability was lower in 1-5 year-old sites when compared to 40-60 year-old sites. I included this hypothesis so that my results were comparable to other amphibian-forestry studies (Petranka et al. 1993; Ash 1997; Knapp et al. 2003). I also hypothesized that salamander occupancy at harvested sites would be positively related to structural elements, such CANOPY and CWD, that were purposefully 152      retained during timber harvest to provide refugia for forest wildlife (Franklin et al. 1997; Rosenvald and Lõhmus 2008). To represent my hypotheses, I constructed models where salamander occupancy (ψ) varied linearly on the logit scale as a function of CANOPY within the 60 x 60m cell (logit(ψi) = β0 + β1(CANOPYi)), or the cumulative number of CWD objects encountered during surveys along the three transects (logit(ψi) = β0 + β1(CWDi)). I hypothesized that CANOPY and CWD could operate independently, or as additive effects on occupancy (ψ(CANOPY + CWD)). Although I stratified canopy cover into different categories during site selection, I treated CANOPY as a continuous variable for all analyses. I predicted that salamander detection probability would be constant across time, as all transects within a site were sampled at the same time and all sites were sampled within a single month. However, I was concerned that variable quantities of downed wood along transects could influence the ability of my field observers to detect salamanders. Hence, I constructed a model where detection probability (p) varied as a function of CWD encountered along transect j (CWDj; logit(pj) = β2 + β3(CWDj). I note the distinction between CWDi, a site covariate, which is the total number of cover objects encountered among all 3 surveyed transects within a site, and CWDj, a detection covariate, which is the number of cover objects along transect j within a site (j = 1,2,3). 5.2.4. Salamander Survival I selected a subset of 6, 1-5 year-old stands that were sampled during my occupancy study. These stands were selected to represent variation in green tree retention levels. I used a 30 x 30m sampling lattice and stratified (i.e., >25%, 10-25%, and <10% canopy cover) random selection to identify 36 cells for study. All cells were >50 m apart. I also selected 9 cells, spaced 153      ≥ 50m apart, within two, 40-60 year-old aspen stands that were adjacent to my 1-5 year-old stands. In early May of 2010 and 2011 I erected a 3 x 3m salamander enclosure at the center of each 30 x 30 m lattice cell (total = 45 cells). Enclosures were constructed of aluminum flashing 50cm high and buried 12-15cm into the ground. The top of each fence was bent inward at 90° to prevent salamander escape. CWD that extended beyond the enclosure boundary were cut and those portions external to the enclosure were removed prior to fence construction. I tested the effectiveness of this enclosure design for preventing salamander escape over the top by adding 2 12 RBSs to a 0.09m replica enclosure, placed in a covered plastic container with air holes, over a 3 day period. No salamanders escaped from the replica during this time. In mid-May, I added 10 adult RBSs that were individually marked with visual implant elastomere (VIE; Northwest Marine Technology, Shaw Island, Washington; Grant 2008) to each enclosure. These salamanders were captured within 1km of my study stands and added to the enclosure within 24 hours of capture. Release locations within an enclosure were randomly 2 assigned. Marked salamander density within the enclosures (≈1.1m ) was comparable to observed densities in mature forests in northern Michigan (Heatwole 1962). In mid-June, I searched enclosures for marked salamanders on three successive visits, separated by 3.0 ± 1.2 days (mean ± 1 standard deviation). Prior to each search, enclosures were gridded into 1m2 sections to ensure searches were performed systematically. Observers thoroughly searched their assigned 1m2 area by examining leaf litter and underneath and within pieces of CWD for marked salamanders. When performing searches, observers placed all leaf2 litter and CWD in plastic tubs that were assigned to each 1m area until nothing remained inside 154      the enclosure except for rooted herbaceous and woody vegetation, and mineral soil. Observers placed leaf-litter and CWD back in the enclosure after searching was complete and attempted to reconstruct the micro-habitat to pre-sampling conditions. All captured salamanders were held in coolers until the sampling event was complete, and then identified by their unique VIE marking, measured, and re-released into the enclosure at the point of capture. I did not mark any unmarked salamanders that were captured inside the enclosures (mean number of captured, unmarked salamanders per enclosure = 4.9). My sampling and handling protocols were approved by the Michigan State University Animal Care and Use Committee (Animal Use Form no. 07/08-118-00). I used the Huggins parameterization of the robust design population model (Pollock 1982, Bailey et al. 2004b) to estimate individual salamander survival (Si), initial capture (pt) and recapture probabilities (ct). Here, I used subscript “i” to denote that survival estimates apply to individual salamanders, as opposed to estimates from occupancy models which apply to individual sites. My robust design framework consisted of two primary periods: a salamander additions period and a capture/recapture period. During the additions period, marked salamanders were added to the enclosure, as described above. After one month, I searched through all enclosures for marked salamanders on three successive visits (i.e., 3 secondary periods during primary period 2). Thus, represents the probability that a marked salamander survived from mid-May (primary period 1) until mid-June (primary period 2) and was available for capture during primary period 2. Initial capture (pt) is the probability that a marked salamander was captured for the first time during visit t of the second primary period (t = 1, 2, 3). Recapture (ct) is the probability a marked salamander was recaptured during visit t, conditional 155      on it being captured at least once before during a previous visit (note: c1 = 0). My design contained only two primary periods, which prevented estimation of emigration between primary periods. I therefore fixed emigration (γ) at 0 for all models. I hypothesized that salamander survival probability in 40-60 year-old sites would be higher than survival in 1-5 year-old sites, given the literature suggesting that amphibians are negatively impacted by timber harvest (e.g., deMaynadier and Hunter 1995; Semlitsch et al. 2009). I also predicted that salamanders within enclosures with higher quantities of CWD would exhibit higher survival rates because these structures trap moisture on the forest floor and may reduce risk of amphibian desiccation (Rittenhouse et al. 2008). I modeled CWD quantity as a continuous covariate on salamander survival (logit (Si) = β0+ β1 × CWDe) where CWDe equaled the total number of CWD objects within the enclosure which contained salamander i. I also hypothesized that salamander survival would be higher within enclosures that were shaded by retained green-trees. To represent this hypotheses, I modeled CANOPY as a continuous covariate on salamander survival (logit(Si) = β2+ β3 × CANOPYe) where CANOPY equaled the amount of canopy cover within the 30 x 30m cell in the enclosure that contained salamander i. As with the occupancy analysis, I modeled survival probability as a function of a single covariate only (S(CANOPY) or S(CWD)), or as an additive effect (S(CANOPY + CWD)). I explored whether capture and recapture probabilities were equal and constant across time (p(.)). Although seldom acknowledged, perfect or constant capture probabilities are often implicitly assumed by investigators who use organism counts to infer amphibian response to an ecological effect (Pellet and Schmidt 2005, Kroll 2009). For contrast, I also hypothesized that capture and recapture probabilities may decrease through time because previous sampling 156      disturbances caused salamanders to be temporarily unavailable for sampling during later surveys (p(t)). This hypothesis is based on previous work which suggests salamanders temporarily emigrate from sites following habitat disturbances caused by sampling (Bailey et al. 2004a; Otto and Roloff 2011). I note the distinction between emigration that occurs between primary periods (“γ” in traditional Robust Design models), which I fixed at “0”, and temporary emigration that occurs within a primary period, between surveys. Although my enclosures restricted horizontal movements of salamanders, temporary unavailability could still occur if salamanders moved vertically into the subterranean environment following a sampling disturbance. My final hypothesis was based on the prediction that salamanders may exhibit a behavioral response to being captured, thereby resulting in lower recapture probability during subsequent surveys (i.e., trap response). I represented this hypothesis by including an additive constant on recapture probability, which allowed initial capture and recapture probabilities to be different (p(.) = c(.)+b; Lukacs 2007). I also considered a similar model where initial capture and recapture probabilities differed and varied across time (p(t) = c(t)+b). 5.2.5 Data Analysis I analyzed the occupancy data using R (version 2.12.1, http://www.r-project.org/; R Development Core Team 2011) with the add-in package unmarked (Fiske and Chandler 2011). I analyzed survival data using Program MARK (MARK, version 5.1, http://www.cnr.colostate.edu/~gwhite/mark/mark.htm). For both studies I used Akaike’s Information Criterion, adjusted for small sample size (AICc), to rank models (Burnham and Anderson 2002). I changed the effective sample size for the survival analysis to represent the number of field enclosures (n= 45), rather than the number of salamanders (n= 450). I used cumulative AIC weights (w+) and evaluation of 95% confidence intervals to determine relative 157      importance of covariates and model parameters. I report model averaged estimates and unconditional 95% confidence intervals for all real parameters. For my occupancy study, I report confidence intervals for all 40-60 year-old sites based on the profile likelihood approach because estimates often resided on the 0-1 boundary (Gimenez et al. 2005). 5.3. Results Covariates CANOPY and CWD were weakly, negatively correlated for my occupancy (df = 132, 2 correlation coefficient = -0.36 (-0.56 – -0.16; 95% CI), R =0.09) and survival (df = 43, -0.63 (2 1.04 – -0.21), R =0.18) studies. 5.3.1. Salamander Occupancy As predicted, occupancy estimates for RBSs were higher for 40-60 year-old sites ( 40-60yr Stands = 1.0, 0.79 – 1.0; 95%CI) when compared to harvested sites ( 1-5yr Stands = 0.57, 0.41 – 0.72). Weights for models that included the effect of CANOPY and CWD on occupancy were 0.32 and 0.38, respectively (Table 1). Although my structural retention covariates, CANOPY and CWD, both showed a positive relationship with salamander occupancy (Figure 5.1), the effect size (i.e., the associated beta estimates) was small and the 95% confidence intervals overlapped zero. For example, salamander occupancy estimates for sites with 50% canopy retention were only 15% higher than sites with 0% canopy retention (Figure 5.1). Collectively, the evidence does not suggest that patterns in salamander occupancy were related to structural retention attributes within my 1-5 year-old clearcut aspen stands. Estimated detection probability for RBSs was <1.0 for all candidate models (range = 0.27 – 0.49). I observed evidence that salamander detection probability was higher along transects that contained higher quantities of downed wood (cumulative AICc weights; w+ = 0.56; Table 158      1). However, the confidence intervals for CWDj overlapped zero for all models that included this covariate and the effect size was small. For example, my top ranking model estimated detection probability as a function of CWDj (Table 1), and the beta parameter associated with CWDj was 0.23 (-0.03 – 0.49; 95%CI). 5.3.2. Salamander Survival Observers took an average of 83 minutes (range = 47-184) to search through an enclosure during a sampling visit. I had 270 detections of marked salamanders during the three visits. Of these detections, 265 were made while searching underneath leaf-litter and CWD, and 5 were from salamanders found in the bottom of the plastic tubs at the end of the sampling event. Model-averaged salamander survival estimates were 0.62 (0.50 – 0.72) and 0.49 (0.36 – 0.63) for 1-5 and 40-60 year-old sites, respectively. Substantial evidence (w+ = 0.99 for CANOPY and CWD; Table 2) indicated that green tree retention and CWD both positively influenced salamander survival from mid-May until mid-June in clearcut harvest stands (Figure 5.2A,B,C). For my top-ranking model, S(CANOPY + CWD + Control), p(t), beta parameter estimates were 0.73 (0.27 – 1.19; 95%CI) and 1.01 (0.53 – 1.50) for CANOPY and CWD, respectively (Figure 5.2A,B). In many cases, estimated survival was higher for 1-5 year-old sites that contained high levels of canopy cover or CWD (Figure 5.2C) compared to an ‘average’ 4060 year-old site ( 40-60yr Stands = 0.49, 0.36 – 0.63; 95%CI). Survival models that assumed constant detection probabilities of RBSs were not supported (w+ = 0.00; Table 2). Rather, initial capture probabilities of marked salamanders were approximately 50% lower in the second and third site visits when compared to the first visit (Figure 5.3). The estimated probability of capturing a marked salamander at least once during 159      the 3 visits was 0.78 (i.e., 1 ∏3 1 1 ). Model-averaged recapture probabilities were slightly lower than initial capture probabilities (Figure 5.3), but the effect was not strong as models with the capture probability structure p(t)=c(t)+b garnered only 0.30 of the cumulative model weight (w+ = 0.30; Table 2). 5.4. Discussion Ultimately, population processes such as survival, reproduction, and movement determine patterns in species occupancy, richness, and abundance (Williams et al. 2002). Response variables like occupancy and abundance may not be effective for determining faunal response to habitat alteration because they are indirect representations of demographic processes (Van Horne 1983; Yoccoz et al. 2001; Williams et al. 2002). However, an appealing characteristic of occupancy and abundance studies is that they can be conducted over broader spatial and shorter temporal scales, thereby increasing the breadth of inference and applicability to management. My inferences regarding RBS response to structural retention in aspen clearcuts were largely conditional on my state variable of interest. Whereas I found no support for a correlation between structural retention and RBS occupancy, I found strong evidence for a positive effect of retention on individual RBS survival probability. 5.4.1. Does Structural Retention Promote Occupancy? I observed higher estimates of salamander occupancy in 40-60 year-old sites when compared to 1-5 year-old sites, which is consistent with other studies that show patterns in amphibian counts are positively correlated with time since timber harvest (Petranka et al. 1993; Ash 1997; Knapp et al. 2003). Salamander occupancy probability was not strongly related to green-tree or CWD retention level within my young aspen clearcuts. Rather, occupancy estimates for 1-5yr old sites were always 33-48% lower than 40-60 yr old sites, regardless of 160      structure retention levels. I selected occupancy as one of my state variables because a primary goal of structural retention prescriptions is to ensure that faunal populations persist within a forest patch following timber harvest (Franklin et al., 1997; Rosenvald and Lõhmus, 2008). Thus, occupancy, or presence-absence, is proposed as a useful state-variable for assessing wildlife response to this mitigation technique. Furthermore, the spatial extent of my study was consistent with the scale at which forest resources are managed and extracted. This is an important consideration because the spatial scale of past amphibian-forestry studies that assessed occupancy or abundance has often been limited to few experimental forest stands (e.g., Ash 1997, McKenny et al. 2006, Patrick et al. 2006, Todd and Rothermel 2006). Inferring broadscale impacts based on limited spatial extent and site replication is a common issue in research on the effects of timber harvest on forest diversity (Bennett and Adams 2004), even though the potential pitfalls of doing so in amphibian-habitat research have been discussed (Dodd and Dorazio 2004; Mazerolle et al. 2007; Kroll 2009). 5.4.2. Does Structural Retention Promote Survival? In contrast to my occupancy results, I found that retention of vertical green-trees and coarse woody debris had a strong effect on salamander survival in young aspen clearcuts. Past research has shown that clearcutting negatively impacts amphibian survival (Rothermel and Semlitsch 2006; Todd and Rothermel 2006; Rittenhouse et al. 2008). However, the role of structural retention for reducing these impacts is less clear (Rosenvald and Lõhmus 2008). An enclosure study by Rittenhouse et al. (2008) showed that amphibian survival over 30 hours in clearcuts was higher when enclosures were deployed in brushpiles compared to enclosures deployed in open areas with no microhabitat refugia. My study differs from Rittenhouse et al. 2 2 (2008) in both the area encompassed by the enclosure (0.07m ; Rittenhouse et al. 2008, 9m ; my 161      study) and the time over which survival was estimated (30 hours; Rittenhouse et al. 2008, ≈1 month; my study). My research shows that leaving structural elements, such as vertical green trees and horizontal CWD, within young clearcuts may ameliorate the observed negative impacts of clearcutting on amphibian survival (Rothermel and Semlitsch 2006; Todd and Rothermel 2006; Rittenhouse et al. 2008). Various combinations of retention can be used to yield comparable survival estimates, ranging from an emphasis solely on CWD or green trees to a combination of CWD and green trees (Figure 5.2C). 5.4.3. Probability of Encounter My analyses highlight the importance of accounting for imperfect detection probability when assessing the effects of anthropogenic disturbance on amphibian populations; salamander detection probabilities at both the species and individual level were always <1.0, sometimes <0.3. This finding is significant considering count data, which are uncorrected for detection probability, are commonly used to infer amphibian response to timber harvest and other anthropogenic disturbances (e.g., deMaynadier and Hunter 1995, Todd and Rothermel 2006, Homyack and Haas 2009). The importance of accounting for bias caused by limited detectability of target species has been emphasized (MacKenzie and Kendall 2002, Bailey et al. 2004a, Kéry and Schmidt 2008); however, ecologists have been slow to apply these analytical tools to amphibian research (Dodd and Dorazio 2004, Mazerolle et al. 2007, Kroll 2009). Indeed, results from my survival analysis suggest that raw counts of recaptured individuals are likely to underestimate true survival because marked individuals went undetected during sampling events. Even though I implemented several design-based approaches for increasing capture rates (i.e., housing salamanders in enclosures, repeated and intensive sampling), my survival analysis revealed that capture and recapture probabilities were surprisingly low. Thus, my study does not 162      support the notion that design-based approaches alone can eliminate bias caused by limited detectability of amphibians. 5.4.4. Conclusions Single pattern- or process-based response variables are often used to assess the impacts of anthropogenic disturbance on biological diversity; seldom are they jointly evaluated in the same ecological system (Todd and Rothermel 2006, Welsh et al. 2008). Inferential power of studies that directly assess demographic processes, such as survival, is generally greater than studies that assess demographic indices, such as occupancy or abundance; however, strength typically comes at the expense of inferential breadth. Ecologists often fail to recognize that understanding demographic processes at a small number of sites is of little utility to resource managers concerned with broad-scale anthropogenic impacts (Yoccoz et al. 2001). Similarly, managers do not often acknowledge the importance of understanding the underlying demographic processes that result in changes in population indices over space and time. My research shows that structural retention harvest, when done at relatively small spatial scales, can reduce the negative impacts of clearcutting on terrestrial salamanders by lowering mortality rates. This provides resource managers with direct evidence that alternative forest management can positively influence salamander population dynamics. However, when viewed at larger spatial scales (i.e., species occupancy across a landscape), I found little evidence that structural retention resulted in higher site occupancy by RBSs, at least within the first 5 years following timber harvest. These seemingly conflicting results may highlight the scale-dependent nature of my two response variables: ecological processes that drive patterns of salamander occurrence likely operate at spatial scales beyond those observed in this study. Indeed, the composition and configuration of the landscape matrix may be an equal or better predictor of 163      amphibian patch occupancy than within-patch habitat conditions (e.g., Joly et al. 2001, Marsh and Trenham 2001, Otto et al. 2007). Although species occupancy is often used as the response variable in wildlife-forestry research (e.g., Rosenvald and Lõhmus 2008, Welsh et al. 2008, Kroll et al. 2008), my study suggests that the spatial resolution of this variable may be too coarse to effectively assess the impact of small-scale mitigation techniques, such as structural retention, on wildlife populations. Another potential explanation for these conflicting results is the difference in the temporal scales of both studies. My occupancy study represents a snapshot of a population that was subjected to an anthropogenic disturbance event that occurred 1to 5 years previous, whereas my survival study was conducted over just one month. Thus, although structural retention may increase salamander survival over one month, this benefit may be negated over a longer temporal period in clearcuts. One final compelling explanation for my conflicting results is that patterns in salamander occurrence within clearcuts were largely influenced by population movement dynamics, rather than mortality. Some amphibians will emigrate from (i.e., evacuation hypothesis; Semlitsch et al. 2008, Peterman et al. 2011) or may go deep underground (i.e., subterranean retreat hypothesis; Petranka et al. 1993) following timber harvesting events. Thus, lower occupancy estimates for my 1 to 5 year stands could reflect behavioral avoidance of salamanders to clearcuts, regardless of the local availability of structural retention. However, in my enclosure experiment, I restricted the horizontal movements of salamanders; thereby eliminating their ability to seek refugia in adjacent forest stands and instead, forced them to select microhabitat refugia within the enclosure. This explanation is consistent with the observed weak correlation between salamander occupancy and structural retention and the strong effect of structural 164      retention on salamander survival. Although my enclosure experiment shows that salamanders can survive in clearcuts when habitat refugia are provided, it is unclear if these salamanders would choose to remain in clearcuts, if given the option (Semlitsch et al. 2008, Rittenhouse et al. 2009, Todd et al. 2009). My mark-recapture study provides anecdotal evidence that salamanders will make vertical movements into the subterranean environment to escape anthropogenic disturbances to the forest floor: estimated capture probabilities were lower for the second and third enclosure sampling events than the first. Regardless of potential mechanism(s), my study suggests that broad-scale conservation goals for terrestrial salamanders, and other forest-obligate species, will not be accomplished by simply retaining structure within individual harvest units. Rather, successful conservation efforts for these species may also require conservation of late-successional forests as part of managed landscapes. In the foreseeable future, habitat degradation will continue to be a dominant force in the global biodiversity and sustainability crisis (e.g., Stuart et al. 2004; Schipper et al. 2008). Hence, developing research and monitoring programs that assess broadscale changes in population patterns, and the demographic processes underlying these changes, should be a shared goal of ecologists and resource managers alike. 5.5. Acknowledgements I thank D. Burt, A. Coleman, J.D. and J.M. Lautenbach, D. McNeil, R. Sterling, S. Wieferich, and numerous volunteers for their assistance in the collection of field and GIS data. M. Donovan, K. Fitzpatrick, and L. Bailey helped develop project objectives and provided guidance on the analysis. A.J. Kroll and L. Bailey provided thoughtful comments on an earlier version of this manuscript. Support for this project was provided by the MDNR–Wildlife Division with funds from the federal Pittman-Robertson Wildlife Restoration Act grant administered by the 165      United States Fish and Wildlife Service (W-147-R: Michigan’s Statewide Wildlife Research and Restoration Program), the Rocky Mountain Goat Foundation, and the Michigan Society of Herpetologists. 166      APPENDIX 167      Appendix C.1 Methodology for calculating the proportion of canopy cover at study site. To estimate the proportion of canopy cover of each 50m radius site, I digitized canopy cover of all retained green-trees within the harvest units using 2010, 2009, 2005, and 1998 NAIP (National Agricultural Imagery Program) imagery. The age of the harvest unit dictated which year of NAIP imagery I used. For example, if harvest of a stand was completed in 2003, then I used 2005 NAIP imagery to map retention. To digitize canopy cover, I created a new shapefile in a Geographic Information System (GIS; ArcGIS 9.1; Environmental Systems Research Institute, Redlands, CA) to represent canopy cover provided by retained green-trees. I used the SKETCH tool in ArcEditor to draw the canopy cover polygons into the new shapefile. I then used the INTERSECT function in Arc Toolbox (ArcGIS 9.1; Environmental Systems Research Institute, Redlands, CA) to intersect the canopy cover polygon shapefile with separate polygon shapefile that represented the boundary of 50m radius site. I used the AREA function in HAWTH’S TOOLS (Hawth’s Tools, version 3.27, http://www.spatialecology.com/htools/) to calculate the proportion of canopy cover within each site. For example, if area of canopy cover 2 2 within a site was 1260m then the proportion of canopy cover at that site was {1260m / [ π × 2 (50m radius) ]} or 0.16. I used this value as continuous site covariate in my Chapter 5 analysis (i.e., CANOPY). 168      Table 5.1. Ranking of candidate occupancy models for red-backed salamanders detected at 134 sites located in aspen stands from the northern Lower Peninsula of Michigan, USA, 2010-2011. ψ and p represent site occupancy and detection probability, respectively. All models include a “Control” covariate, noting whether a site was located in a recently harvested (1-5 years postharvest) or 40-60 year-old aspen stand. CANOPY represents the green-tree retention covariate, while CWDi and CWDj represent coarse woody debris covariates for each site and transect within a site, respectively. Δ AICc Model a w K −2 l ψ(Control),p(CWDj) ψ(Control),p(.) ψ(CWDi + Control),p(.) ψ(CANOPY + Control),p(CWDj) ψ(CWDi + Control),p(CWDj) ψ(CANOPY + Control),p(.) ψ(CANOPY + CWDi + Control),p(.) 0.00 0.91 0.26 0.16 4 3 394.9 397.9 1.37 1.53 0.13 0.12 4 5 396.3 394.2 1.53 2.55 0.12 0.07 5 4 394.3 397.4 2.77 0.07 5 395.5 ψ(CANOPY + CWDi + Control),p(CWDj) 3.00 0.06 6 393.5 a ΔAICc = difference from the Akaike’s Information Criterion best model, adjusted for small sample size, K = number of parameters, −2 l is twice the negative log-likelihood. 169      Table 5.2. Ranking of candidate robust design models for 450 red-backed salamanders within field enclosures deployed in aspen stands, Lower Peninsula of Michigan, 2010-2011. All models include a “Control” covariate, denoting whether the salamander was detected in 1-5 or 40-60 year-old sites. “CANOPY” represents the amount of green-tree canopy cover above each salamander enclosure, while “CWD” represents the number of coarse woody debris objects within each enclosure. p and c represent initial capture and recapture probability of marked salamanders, respectively. b is an additive constant for recapture probability. Δ AICc Model S(CWD + CANOPY + Control), p(t) S(CWD + CANOPY + Control), p(t) = c(t) + b S(CWD + Control), p(t) S(CWD + CANOPY + Control), p(.) = c(.) + b S(CWD + Control), p(t) = c(t) + b S(CWD + Control), p(.) = c(.) + b S(Control), p(t) = c(t) + b S(CANOPY + Control), p(t) S(Control), p(t) S(CANOPY + Control), p(t) = c(t) + b S(CANOPY + Control), p(.) = c(.) + b S(CWD + CANOPY + Control), p(.) S(CWD + Control), p(.) S(CANOPY + Control), p(.) S(Control), p(.) a 0.0 1.6 9.0 9.7 11.6 22.9 27.8 28.7 29.0 31.4 42.7 73.5 82.6 100.0 100.2 a w K −2 l 0.68 0.30 0.01 0 0 0 0 0 0 0 0 0 0 0 0 7 8 6 6 7 5 5 6 5 7 5 5 4 4 3 1219.8 1217.6 1230.8 1231.5 1230.6 1247.4 1252.3 1250.5 1253.5 1250.3 1267.2 1297.9 1309.6 1327.0 1329.6 ΔAICc = difference from the Akaike’s Information Criterion best model, adjusted for small sample size: K = no. of parameters; −2 l is twice the negative log-likelihood. 170      Figure 5.1. Occupancy probability for red-backed salamanders in 1-5 year-old clearcut aspen stands in the northern Lower Peninsula of Michigan, 2010-2011, explained as a function of (A) percent of site area that was covered by green tree canopy retention, and (B) the number of coarse woody debris objects at each site. All estimates were generated using model averaging. A 1 Occupancy Probability (95%CI) 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 20 40 60 Green Tree Canopy Cover (%) 171    80 100   Figure 5.1. (cont’d) B 1 Occupancy Probability (95%CI) 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 20 50 80 110 140 170 Number of Coarse Woody Debris Objects/Site 172      Figure 5.2. Estimated salamander survival probability in the northern Lower Peninsula of Michigan, 2010-2011, explained as a function of (A) percent residual canopy cover found above a field enclosure, (B) the number of coarse woody debris objects within a field enclosure, (C) or both variables. Survival is the probability that a marked salamander i survived 1 month in a field enclosure located in an aspen stand that was 1-5 years post harvest. All estimates were generated using model averaging. A Survival Probability (95% CI) 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 0 10 20 30 40 50 60 70 Green-tree Canopy Cover (%) 173    80 90 100   Figure 5.2. (cont’d) B 1.0 0.9 Survival Probability (95% CI) 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 1 10 19 28 37 45 54 63 72 Number of Cover Objects 174    81 90 99 108   Figure 5.2 (cont’d) C 1 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 90 75 60 45 0 15 30 30 45 60 175    15 75 90 105 0 Green Tree Canopy Cover (%) Survival Probability 0.9   Figure 5.3. Model-averaged estimates of initial capture (p) or recapture (c) probability of salamanders during 3 sampling events (visits) in mid-June, northern Lower Peninsula of Michigan, 2010-2011. Initial capture (pt) is the probability that a marked salamander is captured for the first time during visit t, conditional on it surviving and being available for capture. Recapture (ct) is the probability a marked salamander is recaptured during visit t, conditional on Capture/Recapture Probability (95%CI) it being captured at least once before during a previous visit. 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 p1 Visit 1 p2 Visit 2 c2 p3 c3 Visit 3 176      LITERATURE CITED 177      LITERATURE CITED Archaux, F., Henry, P.Y., Gimenez, O., 2012. When can we ignore the problem of imperfect detection in comparative studies? Methods in Ecology and Evolution 3, 188-194. Ash, A.N., 1997. Disappearance and return of plethodontid salamanders to clearcut plots in the southern Blue Ridge Mountains. Conservation Biology 11, 983-989. Bailey, L.L., Simons, T.R., Pollock, K.H. 2004a. Estimating site occupancy and species detection probability parameters for terrestrial salamanders. Ecological Applications 14, 692-702. Bailey, L.L., Simons, T.R., Pollock, K.H., 2004b. Comparing population size estimators for plethodontid salamanders. Journal of Herpetology 38, 370-380. Bennett, L.T., Adams, M.A., 2004. Assessment of ecological effects due to forest harvesting: approaches and statistical issues. Journal of Applied Ecology 41, 585-598. Bielecki, J., J. Ferris, K. Kintigh, Koss, M., Kuhr, D., MacKinnon, S., Throop, S., Visser, L., Walters, M., 2006. Within stand retention guidance. Michigan Department of Natural Resources, Lansing, MI. http://www.michigan.gov/dnr/. Burnham, K. P., and D. R. Anderson. 2002. Model Selection and Multi-model Inference: A Practical Information-theoretic Approach. 2nd edn. Springer-Verlag, New York, New York, USA. Carlson, A., Edenhamn, P., 2000. Extinction dynamics and the regional persistence of a tree frog metapopulation. Proceedings of the Royal Society of London Series B-Biological Sciences 267, 1311-1313. Cushman, S.A., 2006. Effects of habitat loss and fragmentation on amphibians: A review and prospectus. Biological Conservation 128, 231-240. Davic, R.D., Welsh, H.H., 2004. On the ecological roles of salamanders. Annual Review of Ecology Evolution and Systematics 35, 405-434. deMaynadier, P.G., Hunter M.L., Jr., 1995. The relationship between forest management and amphibian ecology: a review of the North American literature. Environmental Reviews 3, 230-261. Ferraz, G., Russell, G.J., Stouffer, P.C., Bierregaard, R.O., Pimm, S.L., Lovejoy, T.E., 2003. Rates of species loss from Amazonian forest fragments. Proceedings of the National Academy of Sciences of the United States of America 100, 14069-14073. Fiske, I. and R.B. Chandler. 2011. Unmarked: An R package for fitting hierarchical models of wildlife occurrence and abundance. Journal of Statistical Software 43, 1-23. 178      Foley, J.A., DeFries, R., Asner, G.P., Barford, C., Bonan, G., Carpenter, S.R., Chapin, F.S., Coe, M.T., Daily, G.C., Gibbs, H.K., Helkowski, J.H., Holloway, T., Howard, E.A., Kucharik, C.J., Monfreda, C., Patz, J.A., Prentice, I.C., Ramankutty, N., Snyder, P.K., 2005. Global consequences of land use. Science 309, 570-574. Franklin, J.F., Berg, D.R., Thornburgh, D.A., Tappeiner, J.C., 1997. Alternative silvicultural approaches to timber harvesting: Variable retention harvest systems, in: Kohm, K.A., Franklin, J.F. (Eds.), Creating a Forestry for the 21st Century: The Science of Ecosystem Management. Island Press, Washington D.C. pp. 111-138. Gibson, L., Lee, T.M., Koh, L.P., Brook, B.W., Gardner, T.A., Barlow, J., Peres, C.A., Bradshaw, C.J.A., Laurance, W.F., Lovejoy, T.E., Sodhi, N.S., 2011. Primary forests are irreplaceable for sustaining tropical biodiversity. Nature 478, 378-381. Gimenez, O., Choquet, R., Lamor, L., Scofield, P., Fletcher, D., Lebreton, J.D., Pradel, R., 2005. Efficient profile-likelihood confidence intervals for capture-recapture models. Journal of Agricultural Biological and Environmental Statistics 10, 184-196. Grant, E.H.C., 2008. Visual implant elastomer mark retention through metamorphosis in amphibian larvae. Journal of Wildlife Management 72, 1247-1252. Heard, G.W., Scroggie, M.P., Malone, B.S., 2012. Classical metapopulation theory as a useful paradigm for the conservation of an endangered amphibian. Biological Conservation 148, 156-166. Heatwole, H. 1962. Environmental factors influencing local distribution and activity of the salamander, Plethodon cinereus. Ecology 43:460-472. Hof, C., Araujo, M.B., Jetz, W., Rahbek, C., 2011. Additive threats from pathogens, climate and land-use change for global amphibian diversity. Nature 480, 516-519. Homyack, J.A., Haas, C.A., 2009. Long-term effects of experimental forest harvesting on abundance and reproductive demography of terrestrial salamanders. Biological Conservation 142, 110-121. Joly, P., C. Miaud, A. Lehmann, and O. Grolet, 2001. Habitat Matrix Effects Pond Occupancy in Newts. Conservation Biology 15, 239-248. Kendall, W.L., White, G.C., 2009. A cautionary note on substituting spatial subunits for repeated temporal sampling in studies of site occupancy. Journal of Applied Ecology 46, 11821188. Kéry, M., Schmidt, B.R., 2008. Imperfect detection and its consequences for monitoring for conservation. Community Ecology 9, 207-216. Knapp, S.M., Haas, C.A., Harpole, D.N., Kirkpatrick, R.L., 2003. Initial effects of clearcutting and alternative silvicultural practices on terrestrial salamander abundance. Conservation Biology 17, 752-762. 179      Krauss, J., Bommarco, R., Guardiola, M., Heikkinen, R.K., Helm, A., Kuussaari, M., Lindborg, R., Ockinger, E., Partel, M., Pino, J., Poyry, J., Raatikainen, K.M., Sang, A., Stefanescu, C., Teder, T., Zobel, M., Steffan-Dewenter, I., 2010. Habitat fragmentation causes immediate and time-delayed biodiversity loss at different trophic levels. Ecology Letters 13, 597-605. Kroll, A.J., Risenhoover, K., McBride, T., Beach, E., Kernohan, B.J., Light, J., Bach, J., 2008. Factors influencing stream occupancy and detection probability parameters of streamassociated amphibians in commercial forests of Oregon and Washington, USA. Forest Ecology and Management 255, 3726-3735. Kroll, A.J., 2009. Sources of uncertainty in stream-associated amphibian ecology and responses to forest management in the Pacific Northwest, USA: A review. Forest Ecology and Management 257, 1188-1199. Kroll, A.J., Runge, J.P., MacCracken, J.G., 2009. Unreliable amphibian population metrics may obfuscate more than they reveal. Biological Conservation 142, 2802-2806. Lukacs, P.M., 2007. Closed population capture-recapture models. In: Program Mark: a gentle introduction. (eds.) Cooch, E. and White, G.C. [http://www.phidot.org/software/mark/docs/book/] p. 460. MacKenzie, D.I., Kendall, W.L., 2002. How should detection probability be incorporated into estimates of relative abundance? Ecology 83, 2387-2393. MacKenzie, D.I., Nichols, J.D., Lachman, G.B., Droege, S., Royle, J.A., Langtimm, C.A., 2002. Estimating site occupancy rates when detection probabilities are less than one. Ecology 83, 2248-2255. Marsh, D.M., and P.C. Trenham, 2001. Metapopulation dynamics and amphibian conservation. Conservation Biology 15, 40-49. Mazerolle, M.J., Desrochers, A., Rochefort, L., 2005. Landscape characteristics influence pond occupancy by frogs after accounting for detectability. Ecological Applications 15, 824834. Mazerolle, M.J., Bailey, L.L., Kendall, W.L., Royle, J.A., Converse, S.J., Nichols, J.D., 2007. Making great leaps forward: Accounting for detectability in herpetological field studies. Journal of Herpetology 41, 672-689. McKenny, H.C., Keeton, W.S., Donovan, T.M., 2006. Effects of structural complexity enhancement on eastern red-backed salamander (Plethodon cinereus) populations in northern hardwood forests. Forest Ecology and Management 230, 186-196. Otto, C.R.V., Forester, D.C., Snodgrass, J.W., 2007. Influences of wetland and landscape characteristics on the distribution of carpenter frogs. Wetlands 27, 261-269. Otto, C.R.V., Roloff, G.J., 2011. Using multiple methods to assess detection probabilities of forest-floor wildlife. Journal of Wildlife Management 75, 423-431. 180      Patrick, D.A., Hunter, M.L., Calhoun, A.J.K., 2006. Effects of experimental forestry treatments on a Maine amphibian community. Forest Ecology and Management 234, 323-332. Pellet, J., and B. R. Schmidt, 2005. Monitoring distributions using call surveys: estimating site occupancy, detection probabilities and inferring absence. Biological Conservation 123, 27-35. Peterman, W.E., Crawford, J.A., Semlitsch, R.D., 2011. Effects of even-aged timber harvest on stream salamanders: Support for the evacuation hypothesis. Forest Ecology and Management 262, 2344-2353. Petranka, J. W. 1998. Salamanders of the United States and Canada. Smithsonian Institution Press, Washington, DC. Petranka, J.W., Eldridge, M.E., Haley, K.E., 1993. Effects of timber harvesting on southern Appalachian salamanders. Conservation Biology 7, 363-377. Pollock, K. H. 1982. A capture-recapture design robust to unequal probability of capture. Journal of Wildlife Management 46, 757-760. Rittenhouse, T.A.G., Harper, E.B., Rehard, L.R., Semlitsch, R.D., 2008. The role of microhabitats in the desiccation and survival of anurans in recently harvested oak-hickory forest. Copeia 2008, 807-814. Rittenhouse, T.A.G., Semlitsch, R.D., Thompson, F.R., 2009. Survival costs associated with wood frog breeding migrations: effects of timber harvest and drought. Ecology 90, 16201630. Rosenvald, R., Lõhmus, A., 2008. For what, when, and where is green-tree retention better than clear-cutting? A review of the biodiversity aspects. Forest Ecology and Management 255, 1-15. Rothermel, B.B., Semlitsch, R.D., 2006. Consequences of forest fragmentation for juvenile survival in spotted (Ambystoma maculatum) and marbled (Ambystoma opacum) salamanders. Canadian Journal of Zoology 84, 797-807. Schipper, J., et al., 2008. The status of the world's land and marine mammals: Diversity, threat, and knowledge. Science 322, 225-230. Semlitsch, R.D., Conner, C.A., Hocking, D.J., Rittenhouse, T.A.G., Harper, E.B., 2008. Effects of timber harvesting on pond-breeding amphibian persistence: Testing the evacuation hypothesis. Ecological Applications 18, 283-289. Semlitsch, R.D., Todd, B.D., Blomquist, S.M., Calhoun, A.J.K., Gibbons, J.W., Gibbs, J.P., Graeter, G.J., Harper, E.B., Hocking, D.J., Hunter, M.L., Patrick, D.A., Rittenhouse, T.A.G., Rothermel, B.B., 2009. Effects of timber harvest on amphibian populations: understanding mechanisms from forest experiments. Bioscience 59, 853-862. 181      Stuart, S.N., Chanson, J.S., Cox, N.A., Young, B.E., Rodrigues, A.S.L., Fischman, D.L., Waller, R.W., 2004. Status and trends of amphibian declines and extinctions worldwide. Science 306, 1783-1786. Todd, B.D., Rothermel, B.B., 2006. Assessing quality of clearcut habitats for amphibians: Effects on abundances versus vital rates in the southern toad (Bufo terrestris). Biological Conservation 133, 178-185. Van Horne, P. 1983. Density as a misleading indicator of habitat quality. Journal of Wildlife Management. 47, 893-901. Wake, D.B., Vredenburg, V.T., 2008. Are we in the midst of the sixth mass extinction? A view from the world of amphibians. Proceedings of the National Academy of Sciences. 105, 11466-11473. Welsh, H.H., Pope, K.L., Wheeler, C.A., 2008. Using multiple metrics to assess the effects of forest succession on population status: A comparative study of two terrestrial salamanders in the US Pacific Northwest. Biological Conservation 141, 1149-1160. Williams, B. K., Nichols, J. D., and M. J. Conroy. 2002. Analysis and Management of Animal Populations. Academic Press, San Diego, CA. Yoccoz, N.G., Nichols, J.D., Boulinier, T., 2001. Monitoring of biological diversity in space and time. Trends in Ecology & Evolution 16, 446-453. 182      CONCLUSIONS The overarching theme of my dissertation involves the development and implementation of a conservation research program for studying the impacts of anthropogenic disturbances, specifically timber harvesting, on forest wildlife. Although the primary objective of my dissertation is to assess the effectiveness of structural retention (i.e., canopy green trees and coarse woody debris) for conserving wildlife in Michigan’s harvest forests, it also provides valuable insight to other areas of ecology and wildlife management. A strength of my dissertation is that it shows how pilot data can be used to design future wildlife research projects with strong inferential power and reduced estimation bias. It also highlights how statistical models can be used to decouple biological process from observational sampling error: a problem that is ubiquitous in observational wildlife studies. My capstone chapter (Chapter 5) serves as a cautionary reminder that scientific inference is conditional on the selected response variable(s) and stresses that joint evaluation of population indices, such as species occupancy, and population processes, such as survival, provides a more holistic view of biological reality. In Chapter 1 I used species occupancy models and multiple trapping methodologies to estimate detection probabilities of forest-floor wildlife. I used parameter estimates from my initial analysis to explore alternative study design tradeoffs for sampling two shrew species. Although I captured 25 species, I could reliably detect (detection probability >0.15) only northern short-tailed shrews (Blarina brevicauda) and pygmy and masked shrews (Sorex spp.) using drift fences and red-backed salamanders (Plethodon cinereus) using visual encounter surveys (VES). This research demonstrated the need to incorporate species detectability when comparing the effectiveness of different trapping methodologies and highlights the utility of power analyses for exploring study design tradeoffs for research and monitoring programs. 183      In Chapter 2 I compared the effectiveness of natural cover object (NCO) and leaf litter surveys for detecting red-backed salamanders in harvested forests. Both NCO and leaf litter surveys performed well in their ability to detect P. cinereus and produced similar site occupancy estimates; however, leaf litter searches took 3 times longer to complete. My simulation study revealed the time required to complete leaf litter surveys limited the number of sampling sites, which reduced statistical power for detecting a 25% change in salamander occupancy between two groups of sites. Chapters 1 and 2 illustrate how pilot data can be used to design occupancy studies that possess strong inferential power. Analyses from Chapter 3 revealed general sensitivity of estimates from single-season occupancy models to violations of closure. I observed a chronic decrease in salamander occupancy associated with cumulative sampling and non-random salamander movements over the entire field season. My simulations showed that estimation bias was minimal when extinction probability or the number of sample occasions was relatively low. This research highlights the importance of addressing closure in occupancy studies. I provide multiple solutions, using both design- and model-based frameworks, for minimizing bias associated with non-random changes in occupancy and repeated sampling disturbances. This research should appeal to researchers who are currently designing occupancy studies and are concerned if their design is consistent with occupancy model assumptions. Results from Chapter 3 also provided guidance for development of candidate models and statistical analyses for Chapters 4 and 5. In Chapter 4 I evaluated the effectiveness of green-tree retention for promoting occupancy of interior forest, generalist, and early-successional songbirds in aspen stands that were 1 to 15 years post-harvest. I estimated songbird occurrence at 222 sites where green tree retention was exceeded, met, or below recommended retention guidelines established by the 184      MDNR. Contrary to my initial expectation, results suggest that current green-tree retention guidelines are ineffective at promoting site occupancy of most forest songbirds in aspen harvest units. Rather, site occupancy probabilities for most forest birds showed a clear association with age since harvest (1-15 years); however, the strength and direction of the association varied by species. My research suggests it is unlikely that management goals for many interior forest songbirds will be accomplished solely through green-tree retention at the forest stand scale. I provide natural resource managers with recommendations based on alternative management objectives for forest songbirds. Future management objectives in aspen forests should reflect the value of clearcutting to early-successional and generalist bird species, while recognizing that green-tree retention as measured in my study does little to promote occupancy of the interior forest bird community. Chapter 5 highlighted the importance of coupling large-scale distributional studies with more focused demographic research. Inferential power of studies that directly assess demographic processes, such as survival, is generally greater than the inferential power from studies that assess population indices, such as occupancy; however, strength typically comes at the expense of inferential breadth. My research shows that structural retention harvest, when evaluated at relatively small spatial scales, reduces the negative impacts of clearcutting on terrestrial salamanders by lowering mortality rates. This provides resource managers with direct evidence that alternative forest management can positively influence salamander population dynamics. However, when viewed at larger spatial scales (i.e., species occupancy across a landscape), I found little evidence that structural retention resulted in higher site occupancy by red-backed salamanders, at least within the first 5 years following timber harvest. Thus, results from Chapter 5 corroborate those of Chapter 4, which suggested that broad-scale conservation 185      goals for terrestrial salamanders, and other forest-obligate species, will not be accomplished by simply retaining structure within individual harvest units. Rather, successful conservation efforts for these species likely require conservation of late-successional forests as part of managed landscapes. In the foreseeable future, habitat degradation will continue to be a dominant force in the global biodiversity and sustainability crisis. Hence, developing research and monitoring programs that assess broad-scale changes in population patterns, and the demographic processes underlying these changes, should be a shared goal of ecologists and resource managers alike. 186