A Century of Avian Community Change in the Desert Southwest

By

Kelly J. Iknayan

A dissertation in partial satisfaction of the

requirements for the degree of

Doctor of Philosophy

in

Environmental Science, Policy, and Management

in the

Graduate Division

of the

University of California, Berkeley

Committee in Charge:

Professor Steven R. Beissinger, Chair

Professor David D. Ackerly

Professor Justin S. Brashares

Summer 2018

A Century of Avian Community Change in the Desert Southwest ©2018 By Kelly J. Iknayan

1

Abstract A Century of Avian Community Change in the Desert Southwest

By

Kelly J. Iknayan

Doctor of Philosophy in Environmental Science, Policy, and Management

University of California, Berkeley

Professor Steven R. Beissinger, Chair

Species extinctions and population declines have accelerated over recent decades due to habitat destruction, overexploitation, and invasive species, with cascading effects on ecosystem functions and services as well as human well-being. Climate change has emerged as another powerful driver of species decline, one whose effects are beginning to intensify. It should lead to shifts in species distributions and rearranged communities, unless climatic disruption acts as a systemic threat leading to a community collapse. Desert birds comprise a species-rich, easily detectable assem-blage, and are closely coupled to their physical environment, which makes them suitable indicators of climatic change. I assessed how desert bird populations have been impacted over the past cen-tury by resurveying 106 sites throughout the Mojave and Great Basin deserts that were originally surveyed for avian diversity in the early 20th century by Joseph Grinnell and colleagues from the Museum of Vertebrate Zoology at University of California, Berkeley.

Multispecies occupancy models were employed throughout this research to capture the dynamics of the entire avian community. In my first chapter, I review the impacts of imperfect detection on the estimation of community diversity and how the application of multispecies occu-pancy models to estimate these measures can alleviate this source of error. The hierarchical struc-ture of the model allows data from the entire sample to inform the estimation of occupancy, colo-nization, survival, and detection probabilities, despite encounter histories being stratified by his-toric and modern surveys, species, site, and visit. estimation of community-level values, and the structure of the model itself facilitates the modeling of all species, including rare ones. Correcting for detection is particularly important when using historic data, because differences in methodology and changes in technology that can alter the detectability of species through time can also influence the conclusions drawn from the comparison.

Deserts, already defined by climatic extremes, have warmed and dried more than other regions in the contiguous United States due to climate change. In my second chapter, I assessed how climate change and habitat disturbance have impacted bird populations of the Mojave De-serts. The resurveys of sites originally visited in the early 20th century found Mojave Desert birds strongly declined in occupancy and sites lost nearly half of their species. Declines were associated

2

with climate change, particularly decreased precipitation. The magnitude of the decline in the avian community and the absence of species is excep-tional. Our results provide evidence that bird communities in the Mojave Desert have collapsed to a new, lower baseline. Declines could accelerate with future climate change, as this region is pre-dicted to become drier and hotter by the end of the century.

Where the Mojave and Great Basin meet is a juncture of two distinct avifauna. My third chapter uses a dynamic multispecies occupancy model to evaluate the cumulative effects of the redistribution of all 162 observed breeding species across space, time, and biomes. Cross-system comparisons can verify that trends are more than just regional in nature, which legitimizes their application to the development of broad-scale predictive models or management recommenda-tions. The Mojave, a warm desert, and the Great Basin, a cold desert, are two very different systems with distinct vegetation and animal assemblages. The contiguity of the deserts creates an ideal place to study how 20th century climate change is differentially impacting communities across biomes. A transition zone can persist through a changing climate if biotic factors lag behind climate, such as the leading and trailing-edge range disequilibria in vegetation response, and inherent abiotic factors, such as topography and edaphic factors that contribute to the Mojave-Great Basin transi-tion. Barriers to northward expansion for birds could result in range collapse if the southern limits of warm desert species are also contracting. I evaluated whether species of the warmer Mojave, which continues to warm, are expanding into the colder Great Basin Desert, which has incurred less warming, or whether the transition zone behaves as a barrier to northward expansion. Indi-vidual species experienced occupancy changes in consistent directions in both deserts. A substan-tial proportion of species were in decline in one or both deserts (39.5%), while relatively few species were increasing (6.2%). Most species (n = 80) shifted one or both of their latitudinal range limits. Range shifts were highly idiosyncratic in nature, causing the avifaunas of the two deserts to be less strongly structured than they were in the past. The redistribution of species is driving the genesis of novel communities which may have ecological consequences that are yet to be realized.

This dissertation presents a detailed picture of how desert avifauna has change over the past century. The unique ecology of the deserts means that can serve as bellwethers of climate change. Community collapse of the Mojave avifauna and the redistribution of species across both deserts would have been overlooked without the original faunal surveys of Joseph Grinnell and colleagues from the early 20th century. Although similar changes may be occurring in other ecore-gions that lack comparable historical data, the harsh nature of desert environments makes them more likely to become less suitable for life and offers a prescient warning for biodiversity loss as future climates are pushed further toward extremes.

i

For Lizzie

ii

Table of Contents 1 Detecting Diversity: Emerging Methods to Estimate Species Diversity .................................... 1

1.1 Diversity and Imperfect Detection ...................................................................................... 1

1.2 Origins of Imperfect Detection ............................................................................................ 2

1.3 Hierarchical Detection-based, Multispecies Models to Estimate Diversity .................... 5

Box 1.1: Multispecies Models for Estimating Occupancy and Abundance ............ 5

1.4 Richness and Diversity Estimation in the Multispecies Detection-based Modeling Framework .......................................................................................................................... 8

Box 1.2: An Example Illustrating Multispecies Modeling ........................................ 9

1.5 Comparison of Detection-based Estimators of Diversity with Traditional Estimators 11

1.6 Concluding Remarks: The Future of Diversity Metrics................................................... 13

Box 1.3: Assumptions and Limitations of MSOMs ................................................. 14

1.7 Acknowledgements ............................................................................................................. 15

Glossary ....................................................................................................................... 16

2 Collapse of a Desert Bird Community over the Past Century Driven by Climate Change ... 17

2.1 Introduction ......................................................................................................................... 17

2.2 Methods ................................................................................................................................ 22

2.2.1 Historic Surveys ................................................................................................. 22

2.2.2 Modern Survey Methods .................................................................................. 22

2.2.3 Occupancy Model for Community Dynamics ............................................... 23

2.2.4 Model Covariates ............................................................................................... 25

2.2.5 Species Traits as Predictors of Change ............................................................ 26

2.2.6 Diversity and site characteristics ...................................................................... 27

2.3 Results ................................................................................................................................ 27

2.3.1 Change in the Mojave Avian Community over the Past Century ................ 27

2.3.2 Drivers of Mojave Avifauna Change ............................................................... 28

2.4 Discussion............................................................................................................................. 31

2.4.1 Collapse of a Desert Bird Community ............................................................ 31

2.4.2 The Drivers of Collapse ..................................................................................... 33

iii

2.4.3 Winners and losers ............................................................................................ 35

2.4.4 Conservation Implications ............................................................................... 36

2.4.5 Acknowledgements ........................................................................................... 36

3 Making the Transition: Avian Biogeographic Response to Climate Change across Biomes 38

3.1 Introduction ......................................................................................................................... 38

3.2 Methods ................................................................................................................................ 41

3.2.1 Study Area .......................................................................................................... 41

3.2.2 Field Surveys ...................................................................................................... 42

3.2.3 Modeling Framework ........................................................................................ 43

3.2.4 Modeling Latitudinal Shifts .............................................................................. 44

3.2.5 Partitions of Beta Diversity and Transition Zone Dynamics ........................ 44

3.2.6 Elements of Metacommunity Structure .......................................................... 45

3.3 Results ................................................................................................................................ 46

3.3.1 Avifauna and Species-level Changes ................................................................ 46

3.3.2 Shifting Ranges .................................................................................................. 47

3.3.3 Community Turnover and the Transition Zone ............................................ 49

3.3.4 A Shift in Community Structure ...................................................................... 50

3.4 Discussion............................................................................................................................. 51

3.4.1 A Redistribution of Species............................................................................... 51

3.4.2 Communities in Flux......................................................................................... 52

3.4.3 Conservation Implications ............................................................................... 53

References ......................................................................................................................................... 55

Appendix A ....................................................................................................................................... 78

Appendix B ....................................................................................................................................... 91

iv

List of Figures Figure 1.1. Factors Affecting the Probability of Detection of a Species .............................................. 3 Figure 1.2. The Interaction of Imperfect Detection and Different Sampling Strategies ................... 4 Figure 1.3. Examples of Improved Inferences for Species Richness Estimation from MSOMs .... 13 Figure 2.1 Change in Mean Annual Temperature in the Mojave ..................................................... 18 Figure 2.2 Change in Annual Precipitation in the Mojave ................................................................ 19 Figure 2.3 Habitat Quality in the Mojave Desert ................................................................................ 20 Figure 2.4 Study Area and Survey Locations. ...................................................................................... 21 Figure 2.5 Change in Occupancy and Richness in Mojave Avifauna over the Past Century. ........ 27 Figure 2.6 Drivers of Mojave Avian Community Persistence ........................................................... 29 Figure 2.7 Relative Occupancy Change by Taxonomic Family ......................................................... 31 Figure 2.8 Relationship between Species' Traits and Relative Occupancy Change ......................... 33 Figure 2.9 Relationship between Site Characteristics and Relative Richness Change .................... 35 Figure 3.1 Study Area and Survey Locations. ...................................................................................... 39 Figure 3.2 Occupancy Change by Desert ............................................................................................. 46 Figure 3.3 Species' Occupancy Probability by Latitude ...................................................................... 47 Figure 3.4 Latitudinal Range Limit Shifts. ........................................................................................... 48 Figure 3.5 Direction of Latitudinal Shifts ............................................................................................ 49 Figure 3.6 Change in Community Composition and the Ecoregion Transition ............................. 50

v

List of Tables Main Text

Table 2.1 Posteriors of Community Probabilities ............................................................................... 28 Table 2.2 Relationship of Avian Persistence to Covariates ................................................................ 30 Table 2.3 Species Traits and Occupancy Declines .............................................................................. 32 Table 2.4 Site Characteristics and Loss of Richness ............................................................................ 34 Table 3.1. Metacommunity Structure and Associated Statistics. ...................................................... 51

Appendix Table A.1 Species Traits Included in Relative Occupancy Change Comparison. ........................... 78 Table A.2 Relationship of Modeled Probabilities to Covariates ........................................................ 82

vi

Acknowledgements Berkeley has provided an excellent environment for my growth and development as a scientist. I am grateful for having the opportunity to work with my advisor, Steve Beissinger. Steve has been both a mentor and colleague and has been a constant source of encouragement and support. He has been involved in every stage of the process from the cultivation of scattered ideas through the polishing of final products. I would also like to thank my dissertation and qualifying exam com-mittee David Ackerly, Tony Barnosky, Rauri Bowie, Justin Brashares, and Maggi Kelly for their valuable feedback and instruction. I am indebted to my colleagues in the Beissinger lab for their insight and encouragement from the beginning to the end: Karl Berg, Katie LaBarbera, Aline Lee, Juan Li, André M X Lima, Sarah MacLean, Sean Maher, Tierne Nickel, Lindsey Rich, Eric Riddell, Soorim Song, Henry Streby, Corey Tarwater, and Morgan Tingley. A special thanks to Sean Peter-son, Nathan van Schmidt, and Laurie Hall for sharing an office and letting me talk with you and at you over the years. Thanks for all the Oreos.

This research was performed with a tremendous amount of support and resources. If it were not for the hard work, dedication, and foresight of the original surveyors Joseph Grinnell, Jean Linsdale, Alden Miller, and many other scientists my work in the deserts would have no point of comparison. From the pages and pages of detailed notes they left behind, I nearly felt as if I knew them. The Museum of Vertebrate Zoology provided guidance and access to historic field notebooks, maps, and photos. A special thanks to the staff curators, Carla Cicero and Michelle Koo, and the museum archivist, Christina Fidler, for their assistance in navigating the historic ar-chives. This research would not have been completed if it were not for those who assisted in the resurvey effort: Lauren Barth, Thomas Baxter, Chelsea Hawk, and Robert Klotz, all of whom lived out of the back of their trucks roaming the desert for months on end, without a word of complaint. It was a pleasure working with you.

Private landholders and public land managers opened their gates and graciously allowed us to retrace the footsteps of the historic surveyors to access 106 sites across Nevada and California. I am truly amazed by the number of stakeholders who permitted us to work on their land: the National Park Service (Death Valley, Joshua Tree, and Great Basin National Park, and Mojave National Preserve), the National Forest Service (Inyo, San Bernardino, and Humboldt-Toiyabe National Forests), the Fish and Wildlife Service (Ash Meadows Wildlife Refuge and Desert Na-tional Wildlife Range), the Bureau of Land Management, the Nature Conservancy, the Agua Ca-liente Band of Cahuilla Indians, the Nevada Department of Wildlife, and more than a dozen private landholders.

Key financial support was provided by the National Science Foundation and the National Geographic Society. I am also grateful for the travel and research funds provided by: the Joshua Tree National Park Robert Lee Fund; the Ohrbach Foundation; the American Ornithological So-ciety; the Museum of Vertebrate Zoology; the Department of Environmental, Science, Policy, Management; and the Graduate Division.

vii

Finally, graduate school is a long journey that I could not have made without the support of my family and friends. I am fortunate that fate and genetics have bound me to extraordinary set of misfits: my parents, Maggie and Charlie, my sisters, Leah and Robin, and my niece and nephew, Grace and Connor. You all provided the levity and love I needed over the years. Leah, thank you for being my copyeditor, something I know you loved doing. My friends within the department, Briana, Rachel, Jade, and Kristen, kept my sanity by sharing my plight. My friends in the real world, Shehla, Bee, Kathy, and Sam, kept my sanity by maintaining my perspective. And most of all, Jon-athan, you have been there for me through it all.

1

1 Detecting Diversity: Emerging Methods to Estimate Species Diversity1

Estimates of species richness and diversity are central to community and macroecology and are frequently used in conservation planning. Commonly used diversity metrics account for unde-tected species primarily by controlling for sampling effort. Yet the probability of detecting an in-dividual can vary among species, observers, survey methods, and sites. We review emerging meth-ods to estimate alpha, beta, gamma, and metacommunity diversity through hierarchical multi-species occupancy models (MSOMs) and multispecies abundance models (MSAMs) that explicitly incorporate observation error in the detection process for species or individuals. We examine ad-vantages, limitations, and assumptions of these detection-based hierarchical models for estimating species diversity. Accounting for imperfect detection using these approaches has influenced con-clusions of comparative community studies and creates new opportunities for testing theory.

1.1 Diversity and Imperfect Detection Diversity estimates are central to community and macroecology (Brown 1995; Hubbell 2001; Leibold et al. 2004; Harte 2011) and are frequently used in conservation planning as a surrogate for biodiversity and to identify areas in need of protection (Margules and Pressey 2000; Moilanen et al. 2009). Diversity is classically divided into alpha (site-level), beta (turnover across multiple sites), and gamma (composite of all sites in a region) components. The fundamental unit of all diversity metrics is a count of species, individuals, or both. Yet rarely do circumstances occur when all species or all individuals are detected during a survey, regardless of whether the study organisms are birds (Tingley and Beissinger 2013), mammals (Burton et al. 2012), insects (Bried et al. 2012), or plants (Shefferson et al. 2003; Chen et al. 2013). Imperfect detection has predictable conse-quences: when species are common, missed individuals result in underestimation of populations; when species are rare, missed individuals result in false absences. Uncorrected counts of observed species often used in measures of diversity ignore detection altogether and established methods used to account for missed species do not disentangle detection from occurrence.

Hierarchical occupancy models have recently become a standard method to account for false absences when modeling the occurrence of a single species (MacKenzie et al. 2006). These models can distinguish between and elucidate two key processes (Royle and Dorazio 2008): (i) the

1 This work has been published previously and is reproduced here with permission from the pub-lisher, Cell Press: Iknayan KJ, Tingley MW, Furnas BJ, Beissinger SR. 2014. Detecting diversity: Emerging methods to estimate species diversity. Trends in Ecology & Evolution. 29:97 106. doi:10.1016/j.tree.2013.10.012.

2

ecological or state process of interest (e.g., site occupancy); and (ii) the observation process that always accompanies field sampling (e.g., variability in detection of individuals or species) and that makes identification of the state process imperfect. Recent advances have applied these hierarchical

- models (i.e., models that incorporate the underlying detection process sepa-rately from the ecological measure of interest) to surveys comprising multiple species to estimate diversity metrics. They not only account for multiple sources of detection but can estimate the number of species never encountered during surveys (Dorazio and Royle 2005).

In this review, we examine the origins of imperfect detection and discuss how hierarchical detection-based models can be applied to multiple species to improve estimates of diversity. We highlight important advantages, drawbacks, and considerations of using detection-based models for estimating diversity and compare results from this approach with traditional methods. Multi-species hierarchical approaches are relatively new, so both their potential and their limitations have yet to be realized.

1.2 Origins of Imperfect Detection Arriving at valid estimates of species diversity from multispecies surveys requires an appreciation of the myriad ways in which species are imperfectly detected. The detection process can be sepa-rated into three components: individuals emit a signal (either auditory or visual) that is transmitted through the environment and received by an observer. Various factors affect the rate and strength

ceive it. These factors can be decomposed into four groups (Figure 1.1) two that characterize the organism being surveyed (species and individual) and two that relate to sampling logistics (site and survey). Whereas survey- and site-specific factors affecting detectability can be partly ad-dressed through study design, species- and individual-specific differences in detectability can be accounted for only by using statistical methods.

Species vary in their inherent probabilities of detection. Differences in detectability arise volume, movement frequency) and distinctiveness

(e.g., call, size, and color). The rarity of a species also influences detectability and the probability of detecting a species is positively related to its abundance (He and Gaston 2003; Royle and Nichols 2003; Dorazio 2007). Less appreciated is that the local abundance of another species (e.g., a com-petitor or predator) can influence the behavior of the focal species, resulting in either positive or negative changes in detectability (Bailey et al. 2009; Richmond et al. 2010; Waddle et al. 2010). Substantial heterogeneity in detection among individuals can also occur within species. For exam-ple, in species detected by vocalizations, the frequency and volume of calls can differ by body size, sex, and age. Distance from observer affects the probability of detection of individual animals (Buckland et al. 2001) and can also affect plant detections (Buckland et al. 2007; Jensen and Meilby 2012).

Site- and survey-specific detectability factors can influence the detectability of all species

3

(Figure 1.1). The sampling site has intrinsic characteristics that affect detectability relative to com-parable samples at other sites. Site-level heterogeneity in detection derives from factors that im-pede visual or auditory detection regardless of observer, such as habitat structure or noise. Many factors related to the individual survey or sampling event affect detectability. Weather, which gen-erally changes across observation periods, has a strong influence on detectability, affecting both animal behavior and the detection process itself. For example, playback experiments found breezy conditions reduced auditory detections of songbirds by 28% (Simons et al. 2007). Surveys are often conducted by different observers, and observer ability, age, and experience affect detectability (Alldredge et al. 2007; McClintock et al. 2010). These factors can also impact whether or not a species is identified correctly, which can result in false positives. Sampling design (e.g., survey methods, timing, and effort) also affects detection rates. For example, the detectability of animals

Figure 1.1. Factors Affecting the Probability of Detection of a Species Mean detectability can be considered a species-level trait that is a function of behavior, life history, and rarity. In a hierarchical model, species-specific detectability is assumed to come from a common community-level distribution,shown here in purple. A hypothetical distribution of observed detection probabilities for two members of a bird com-munity (indicated by dashed lines and color-coded probability curves) is shown for different aspects affecting the detection process (individual-, site-, and survey-specific factors). As more components of the observation process are considered, variation in the probability of detection increases.

4

often differs by time of day or sampling date (e.g., calling frogs or birds). Although easily con-trolled, sampling design can differ across the course of a study or among studies.

As a result of imperfect detection, the species that occur at a site at any moment in a mul-tisite survey comprise three categories (Figure 1.2A): (i) species detected at the site; (ii) species not detected at the site but detected at other sites; and (iii) species not detected at any surveyed site but which are known to, or could, occur in the metacommunity or regional pool of species, which may or may not be well described. When a site is surveyed only once (Figure 1.2B), the proportion of

Figure 1.2. The Interaction of Imperfect Detection and Different Sampling Strategies Categories of species at surveyed sites resulting from imperfect detection and how they change with different temporal and spatial sampling strategies. (A) The true (unknown) species pool of a metacommunity repre-sented at a site comprises species that have been detected there (green bin), those that have not yet been detected at the site but have been detected at other surveyed sites (yellow bin), and those that have not yet been detected at this or any site but occur in the region (white bin). (B) As temporal and spatial replication (i.e., sampling effort) increases, knowledge of the species pool changes for both the site (green bins) and the metacommunity (green + yellow + white bins). When a site is surveyed few times, the relative size of each bindepends on the actors affecting detectability (Figure 1). If there are few sites and few surveys per site, a large portion of the metacommunity may not be detected (white bin of upper left rectangle), either at the site or at other sites. As the number of surveys per site increases (temporal replication) but not the number of sites surveyed (i.e., little spatial replication), the total number of species detected per site increases (green bin in upper right rectangle), mostly as a result of detecting species that are likely to occur at other sites (yellow bin). When the total number of sites surveyed increases (spatial replication) but not the number of surveys (i.e., little temporal replication), the number of species undetected in the region decreases (white bin in lower left rectangle), but the number of species detected per site remains the same (green bin). As both the number of surveys per site and the number of sites surveyed increase, a greater proportion of species in the metacommu-nity will be known (green + yellow bins in lower right rectangle), either from being directly detected at the site (green bin) or by being detected at other sites (yellow bin).

5

species undetected may be high. Temporal replication improves the resolution of diversity esti-mates for a site. Spatial replication improves the resolution of diversity estimates for the meta-

rarely sufficient to detect all individuals or species, each of the three categories of detected or un-detected species should be modeled to estimate the true richness of a site.

1.3 Hierarchical Detection-based, Multispecies Models to Estimate Diversity Hierarchical detection-based modeling of occupancy or abundance offers an approach for estimat-ing diversity metrics that can incorporate the processes causing imperfect detection (Figure 1.1).

Box 1.1: Multispecies Models for Estimating Occupancy and Abundance The MSOM is an extension of the single-species, single-season occupancy model (Mac-Kenzie et al. 2002) that combines and analyzes the history of detections and non-detec-tions (denoted by 1s and 0s, respectively) of all species encountered during replicated surveys at a set of sites (Dorazio and Royle 2005). The hierarchical model includes three levels, one each for species ( ), site ( ), and replicate ( ). The first level represents the true occurrence states ( ) within the community of all partially observed and never observed species. The second level is the ecological process governing occurrences ( ) of partially observed species. The third level explains the detection history ( ) from the replicated surveys:

superpopulation process (data augmentation); I)

ecological process; II)

observation process. III)

The second and third levels are always present in a single-species model. The MSOM expands these levels across all observed species within the same system of linked, hierar-chical models. It also addresses the community of partially observed and never observed species through addition of the first level. The parameters (proportion of sites occu-pied) and (probability of detection) are the same occupancy and detection probability parameters, respectively, used in a single-species model. The parameter governs the data augmentation variable .

Data augmentation is used in MSOMs to estimate the number of species present in a community (or metacommunity) but not detected at any site. The detection histories of

6

Hierarchical encounter histories nonde-tection at a survey site (Box 1.1). These histories require repeated surveys during a period when the sampled population is assumed to be closed to changes in occupancy. The pattern of detections at occupied sidetections. - occupancy models provide a framework to estimate hierarchically a probability of detection ( ) and a probability of occupancy ( ) over a sampling period using either a frequentist (maximum likelihood estimation) or Bayesian framework (MacKenzie et al. 2006; Kéry and Schaub 2012) predict first, assemble later approach (Ferrier and Guisan 2006) to estimate species richness, where species are modeled individually and richness measures are calculated through aggregation with established metrics (Williams 2009; Ahumada et al. 2011). Although this approach allows occupancy to be modeled with species-specific covari-ates, modeling species individually restricts inferences about diversity to species that have been

observed species are augmented by all-zero detection histories that represent species with unknown identities. The number of all-zero augmented detection histories, , should be arbitrarily large without being computationally unwieldy [see Royle and Dora-zio (2012) for guidance]. An indicator variable, , is modeled such that for species that were either observed (total = ) or unobserved but available for sampling. The total number of unobserved species, , for which , is equivalent to the asymptote of a species-accumulation curve (Kéry and Royle 2009). Consequently, gamma diversity is represented by (where ), which estimates the total number of species in a sample of sites. This can be directly calculated by multiplying the posterior estimate of

by the known quantities . The structure of the MSAM is similar to that of the MSOM, except that the detection

history comprises observed counts ( ) given the true abundances ( ):

superpopulation process (data augmentation); IV)

ecological process; V)

observation process. VI)

Instead of a Poisson distribution controlling the ecological process (Equation V), a zero-inflated Poisson, a negative binomial, or another distribution suitable for counts can be used.

Box 1.1 continued

7

detected multiple times, which excludes species that are rare or are undetected across all sites (Fig-ure 1.2). Additionally, this approach makes it difficult to propagate uncertainty in occurrence es-timates of a single species into uncertainty for community metrics.

As a result of the limitations of aggregating single species occupancy models, estimation of diversity metrics in a detection-based modeli assemble and predict together approach using a hierarchical community occupancy model. We will refer to these as MSOMs when used to estimate incidence-based measures of diversity (Dorazio and Royle 2005; Dorazio et al. 2006; Dorazio et al. 2010). Hierarchical diversity models have been developed with-out explicitly accounting for the detection process (Gelfand et al. 2005; Ovaskainen and Soininen 2010). However, a great advantage of incorporating the detection process into community models is the ability to account explicitly for the effects of survey-, site-, species-, and individual-level fac-tors affecting detectability (Figure 1.1) through the inclusion of one or multiple detection covari-ates (Dorazio and Royle 2005; Dorazio et al. 2006). MSOMs model undetected species in a biolog-ically oriented, process-driven way. The mathematical framework used by MSOMs is discussed in Box 1.1. Although MSOMs can be analyzed using either frequentist or Bayesian methods, current implementations, including available code (Royle and Dorazio 2008; Zipkin et al. 2009), favor the latter.

The hierarchical structure of MSOMs allows data from the entire sample to inform the estimation of diversity, despite encounter histories being stratified by species and site (Dorazio and Royle 2005). The detection and occupancy probabilities of each species are assumed to come from a community-level distribution (e.g., Figure 1.1, bottom panel). For the sake of mathematical con-venience, a normal distribution on the logit scale is often used, with an associated mean and vari ance as hyperparameters. This construction is equivalent to treating the different species as ran-dom effects (Royle and Dorazio 2008; Kéry and Royle 2009; Kéry and Schaub 2012). Community-level hyperparameters facilitate the modeling of all species, including rare ones, through a property often re borrowing strength (Clark et al. 2005) or Bayesian shrinkage (Link and Sauer 1996; Zipkin et al. 2009)variance and, as a result, estimates of individual species are pulled (i.e., shrunk) toward the com-munity mean. Many of the ben , because data are used more efficiently compared with single-species models and individual species estimates are improved (Link and Sauer 1996; Kéry and Schaub 2012). Hierarchical community models have improved the precision of diversity descriptors (Zipkin et al. 2009), which can offset the costs of multispecies monitoring efforts (DeWan and Zipkin 2010). Moreover, borrowing strength allows MSOMs to estimate site-level occupancy for rarely detected species, which is often not possible with single-species models due to the limited number of detections (Kéry and Royle 2008; Dorazio et al. 2011). Although Bayesian shrinkage can improve parameter estimation, estimates for infre-quently encountered species will be pulled more toward the metacommunity mean because they are estimated with less precision and inform the community-level mean less than species that are frequently detected (Ruiz-Gutiérrez et al. 2010; Kéry and Schaub 2012: 80 81).

8

Although most hierarchical detection-based occupancy models can be used to produce es-timates of species richness and turnover based on incidence, the MSOM framework easily accom-modates replacement of species incidence with counts of individuals to create a MSAM to estimate abundance-based diversity metrics (e.g., Shannon 1948; Simpson 1949; Chao et al. 2006). The MSAM is a multispecies version of the N-mixture model that has been used to estimate the abun-dance of individual species (Dodd and Dorazio 2004; Royle 2004; Kéry et al. 2005; Wenger and Freeman 2008; Joseph et al. 2009). It is structurally similar to the MSOM (Box 1.1), but replaces the ecological and observation process in the MSOM with distributions suitable for counts. Com-munity-level hyperparameters of the MSAM represent: (i) the abundance of a species in the met-acommunity; and (ii) the probability of detecting an individual for each species.

Beyond the need for repeated surveys, MSAMs lack the ancillary data requirements char-acteristic of distance sampling and capture recapture methods for estimating abundance (Kéry 2008). They do, however, require assumptions about the probability distribution governing the true number of individuals of each species occurring at each site. Moreover, for each species at a site, the detection of each individual is assumed to be independent of that of other individuals. This can lead to over-inflation of counts and contrasts with the approach taken in distance sampling (Buckland et al. 2001), which models the detectability of groups of individuals. However, a beta-binomial distribution can be applied to address nonindependence of detections of individuals (Martin et al. 2011; Dorazio et al. 2013).

1.4 Richness and Diversity Estimation in the Multispecies Detection-based Modeling Framework MSOMs have been applied to diversity estimation in two situations: (i) when all possible species in the metacommunity or region were detected at least once during sampling, which can be mod-eled with the MSOM structure discussed above (Royle and Dorazio 2008); and (ii) where meta-community or regional species richness exceeds the total number of species detected across all sites (Dorazio et al. 2006). To model richness when the pool of potential species is unknown (a common occurrence), MSOMs employ (Kéry and Royle 2009; Zipkin et al. 2010) to estimate the number of species never detected during sampling in addition to those that were de-tected at least once (Figure 1.2). As discussed in Box 1.1 and illustrated in Box 1.2, this is done by augmenting the detection histories of observed species with an arbitrarily large number of all-zero records that represent the detection histories of hypothetical, unobserved species. The model then estimates how many hypothetical species are likely to occur in the sampling region but were missed by all surveys, in addition to the number of species that were detected at least once [see Royle and Dorazio (2012)] for additional details on data augmentation). The total number of species esti-mated by the data-augmentation process is comparable to the asymptote of the species accumula-tion curve in a homogeneous landscape (Kéry and Royle 2009; Zipkin et al. 2010), but allows for the incorporation of detection and occupancy covariates into the estimation of the asymptote.

9

Thus, data augmentation provides one way to estimate diversity at the scale of the metacommunity (Dorazio et al. 2010) or region (gamma diversity).

Any occurrence-based descriptor of diversity and its associated measures of uncertainty can be derived directly within the MSOM framework for any subset of samples, such as sites,

Box 1.2: An Example Illustrating Multispecies Modeling Here we provide a brief overview of the modeling process used to estimate detection-based community metrics derived from a MSOM developed by Zipkin et al. (2010). Data were collected to determine the effects of two different management treatments on breed-ing-bird diversity. A single-season MSOM was used to account for species-specific dif-ferences in detection and occupancy by site. The model runs using the freely available software R and WinBUGS.

Inputs

Temporally and spatially replicated surveys are required to calculate detection-based community metrics. The number of surveys need not be the same for all sites. One or more explanatory covariates can be incorporated at the species, site, or survey level (e.g., Julian day is a survey-level covariate of detection). Multispecies survey data are organized into a 3D array: survey ( ) site ( ) species ( ). This array includes the encounter his-tories for all observed species. An arbitrarily large number of all-zero encounter histories are appended to this array in a process known as data augmentation (Box 1.1). Continu-ous covariates are typically standardized, which helps with overall MCMC performance and simplifies the comparison of covariate effects when interpreting model output. Model Specification and MCMC Sampling

10

groupings of species, metacommunities, or regions (e.g., Ruiz-Gutiérrez et al. 2010; Burton et al. 2012; Tingley and Beissinger 2013). The multispecies approach retains the identity of encountered

Occupancy ( ) and detectability ( ) parameters can be generalized to vary according to explanatory covariates. This is usually handled as a linear model on the logit scale, where regression coefficients are modeled as species-specific random effects derived from a community-level distribution. Before initiating MCMC sampling, prior distributions are specified for the hyperparameters of the community-level distributions, for the hyperdis-tributions themselves, and for , the parameter that determines membership in the met-acommunity from the superpopulation (Box 1.1). In practice, non-informative priors are often used (Kéry and Royle 2009). Model Output of Posterior Estimates and Summary Analyses

The model returns posterior estimates for species-specific occupancy and detection prob-abilities and the species-specific effects of covariates. Posterior distributions for any de-rived community-level metrics can also be returned. In an MSOM, these metrics can in-clude richness of the metacommunity, richness of individual sites, or richness of sets of sites, as well as richness of different functional groups at those spatial levels. Similarity indices of beta diversity can be also be derived (Tingley and Beissinger 2013). These esti-mates can be used in summary analyses to investigate relationships with covariates or to compare metacommunities.

Box 1.2 continued

11

species throughout the modeling process and allows estimation of species-specific detection and occupancy probabilities, which can be derived from general or group-specific (e.g., taxonomic or functional) covariates (Ruiz-Gutiérrez et al. 2010; Cheal et al. 2012). Currently MSOMs must be written by the user and run within a programming environment or precompiled Markov chain Monte Carlo (MCMC) programs such as WinBUGS (Lunn et al. 2000), JAGS (Plummer 2003), or OpenBUGS (Lunn et al. 2009), all of which use the Bayesian inference Using Gibbs Sampling (BUGS) language. Statistical programs such as R (http://www.R-project.org) can often interface with MCMC programs. Code for various models is available in the BUGS language (e.g., Royle and Dorazio 2008; Kéry and Royle 2009; Zipkin et al. 2010; Dorazio et al. 2011; Tingley and Beissinger 2013).

There are a limited but growing number of studies that have applied MSOMs to estimate diversity. Birds have been the primary target (Russell et al. 2009; Zipkin et al. 2009; Zipkin et al. 2010; Tingley and Beissinger 2013), but applications include amphibians (Walls et al. 2011; Zipkin et al. 2012), freshwater and reef fish (MacNeil et al. 2008; Holtrop et al. 2010; Cheal et al. 2012), and invertebrates (Dorazio et al. 2006; Kéry et al. 2009; Dorazio et al. 2010). MSOM estimation of beta-diversity indices has been done only with the Jaccard and Sørensen indices (Holtrop et al. 2010; Zipkin et al. 2010; Dorazio et al. 2011; Tingley and Beissinger 2013), but the approach could be applied to any of the large number of beta-diversity indices used with occurrence-based data (e.g., complementarity).

Abundance-based estimates of diversity from MSAMs are in their infancy. Only two stud-ies have extended the N-mixture model to a multispecies framework to evaluate communities. Yamaura et al. (2012) modeled community responses of birds and bees to land-use change, using guild-level hyperparameters and data augmentation to differentiate estimated abundances and species richness between early successional and mature forest species. Chandler et al. (2013) mod-eled abundance-based estimates of the Chao-Jaccard similarity index and Shannon diversity to compare the conservation values of competing agricultural systems in a tropical forest landscape. Although diversity estimation with MSAMs has been very limited to date, these models could be used to estimate directly any abundance-based diversity metric, including Hill numbers (Chao et al. 2013), by including them as derived quantities in the model (Box 1.2). An advantage of this approach is that posterior distributions calculated by the MCMC algorithm can be used to describe straightforwardly the uncertainty associated with the estimates of these diversity metrics.

1.5 Comparison of Detection-based Estimators of Diversity with Traditional Estimators Compared with detection-based estimators of diversity, traditional approaches for estimating di-versity have limited ability to incorporate or accommodate the survey-, site-, and species-level pro-cesses that differentially affect the detection of species or individuals (Figure 1.1). The most popu-lar species-richness estimators, the Jackknife (Burnham and Overton 1979) and Chao estimators (Chao 1984; Chao 1987)

12

the proportion of species encountered infrequently to species encountered more frequently to es-timate the number of undetected species (Chao et al. 2005; Mao and Colwell 2005; Dorazio et al. 2006). Although this reduces bias compared with uncorrected species counts, mathematical meth-ods for incorporating additional features of the study design, such as site- or survey-specific co-

etection (Figure 1.1), are not straightforward or often not possible (Kéry and Royle 2009). Beta-diversity indices of community similarity are frequently calculated without correcting for imperfect detection (Chao et al. 2005) and their performance in the face of unde-tected species is variable (Chao et al. 2006; Beck et al. 2013). The Chao-Jaccard/Sørensen index (Chao et al. 2005; Chao et al. 2006) is one of the few beta diversity estimators that corrects for sampling bias by estimating the contribution of undetected species using the probability that indi-viduals drawn randomly from the sample belong to a species shared by the two assemblages, based

rate of species is driven primarily by relative abundance and that detectability does not otherwise differ among observers, species, or sites (Chao et al. 2005), factors that often affect detection (Fig-ure 1.1) and that can be accommodated in MSOMs and MSAMs. Alpha- and gamma-diversity indices, such as the Shannon Wiener and Simpson indices and Hill numbers, which combine rich-ness and relative abundance (Magurran 2004; Ellison 2010), face estimation issues similar to beta diversity. Chao and Shen (2003) and Chao et al. (2013) developed indices and estimators to correct for sampling bias in the Shannon-Wiener index and Hill numbers, respectively. They improve di-versity estimation, but appear unable to correct for most site- and species-level causes of imperfect detection. None of the traditional estimators of diversity is able to differentiate whether the esti-mated number of undetected species represents: (i) species that were detected at other sites; or (ii) species that were not detected in the sample as a whole (Figure 1.2).

Direct comparisons of traditional- and detection-based estimates of species diversity are limited. Two studies compared the second-order Jackknife estimator to a simple MSOM that was built without site- or survey-level covariates (Kéry and Royle 2008; Holtrop et al. 2010). Both mod-els assumed that species identity was the only factor contributing to variation in detectability (Kéry and Royle 2008). MSOM richness estimates were more precise than the second-order Jackknife estimator and generated comparable estimates across surveys with dual observers (Kéry and Royle 2008). Jackknife estimates of richness at the site-level were sometimes erratic, generating reasona-ble estimates for some sites but unreasonably high, inaccurate estimates for others (Kéry and Royle 2008; Holtrop et al. 2010; Figure 1.3A). These differences might reflect more efficient use of data by MSOMs compared with the Jackknife technique. However, comparisons of the methods have not been extensive and this is an area with ample opportunity for further evaluation. Diversity estimates from MSOMs and MSAMs could be compared with estimates from traditional methods using the same simulated and real-world datasets. To date, no study has compared beta-diversity indices of any kind with estimates corrected for heterogeneous detecta-bility derived from a MSOM or MSAM.

We do know, however, that richness estimated with MSOMs and MSAMs is always greater than when estimated from raw species counts (Dorazio et al. 2006), because MSOMs supplement

13

raw species counts with the expected number of undetected species (Kéry and Royle 2009). Using raw counts to measure species diversity can mask temporal trends that are detected with MSOM estimation (Tingley and Beissinger 2013; Figure 1.3B,C) or alter the explanatory power of ecolog-ical covariates (Zipkin et al. 2010).

1.6 Concluding Remarks: The Future of Diversity Metrics Producing accurate estimates of species richness and diversity requires accounting for factors that affect imperfect detection (Figure 1.1) and recognizing the categories of detected and undetected species at surveyed sites (Figure 1.2). Hierarchical multispecies models that incorporate both the detection process and the occurrence state provide a promising way forward, because they lead to a more process-driven estimate of diversity through the delineation of the biological and sampling processes. However, detection-based estimators of diversity generally require multiple surveys at a site to estimate detectability, make assumptions about the closure of populations and about the veracity of species identification (Royle and Link 2006), and are still relatively new so their limita-tions and performance need further evaluation (Box 1.3). Moreover, these hierarchical models are relatively complex and beyond the experience of many potential users. Although the software that runs them is freely available and example code for commonly used models is found on the Internet, more user-friendly implementations of MSOMs and MSAMs would facilitate their use and ac-ceptance. A package that fully implements these models within the R environment would be a strong first step toward this goal. Above and beyond this, the development of specialized software to run these models would increase the ease of implementation and encourage widespread use,

Figure 1.3. Examples of Improved Inferences for Species Richness Estimation from MSOMs (A) Comparison of bird species richness estimated from MSOMs versus the traditional Jackknife estimator. Estimatorsmatch well when richness is low, but Jackknife estimates diverged at higher richness values and are likely to represent an overestimation. Data taken from Table 1 in Kéry and Royle (2009). (B,C) Conclusions about change in species richness over the past century for birds of the Sierra Nevada are reversed when raw species counts are corrected for detectability using a MSOM. Uncorrected counts (B) suggested species richness increased between surveys conductedby Joseph Grinnell and colleagues from 1911 to 1929 and resurveys conducted by Tingley and Beissinger from 2003 to 2009, whereas MSOM estimates (C) accounting for detectability differences showed that richness declined. (B) and(C) adapted from Figure 1 in Tingley and Beissinger (2013).

14

much like EstimateS (http://purl.oclc.org/estimates) did for traditional measures of species diver-sity or Presence (http://www.mbr-pwrc.usgs.gov/software/presence.html) did for single-species occupancy models.

Although the use of detection-based estimators of diversity is increasing, further innova-tions that address various issues already incorporated into single-species occupancy models are needed. As discussed in Box 1.3, this includes correcting for false positives (Royle and Link 2006; Miller et al. 2011) and evaluating model performance and issues relating to closure. Regarding the

Box 1.3: Assumptions and Limitations of MSOMs The main assumptions of occupancy models and MSOMs are: (i) sites are closed to

); (ii) species are correctly identified; and (iii) the probability of detection and occupancy at a site are independ-ent of detection and occupancy at another site (usually assured by a minimum distance between sites).

Closure might be the most commonly violated assumption of occupancy models because species might be unavailable for detection due to daily or seasonal movements. When closure is violated, modeled detectability is biased low and occupancy is over-estimated (Rota et al. 2009). This, in turn, has the potential to inflate richness esti-mates, which might not be problematic if inflation occurs equally across all compared units. If the closure violation varies among species, however, diversity metrics could be differentially biased by species or site. Kéry et al. (2009) developed a MSOM that allows for temporary emigration between sampling periods, which could reduce this bias. The potential consequences of closure violation on species-diversity metrics de-serve further exploration.

With their focus on false negatives, most occupancy models do not model false pos-itives [but see Royle and Link (2006); Aing et al. (2011); Miller et al. (2011)]. Surveys that rely on visual and aural cues are particularly susceptible to misclassification (Si-mons et al. 2007; McClintock et al. 2010) and small errors in classification can lead to large biases in occupancy (Royle and Link 2006; Miller et al. 2011). Distance from ob-server, observer error, and simultaneous vocalizations of multiple species are the lead-ing causes of misclassification. Observer experience and training instructions did not reduce the prevalence of errors (Miller et al. 2011). Models have been proposed to ac-count for both false negatives and false positives (Royle and Link 2006), but issues with parameter identifiability arise when they are applied to data with heterogeneous de-tection (Fitzpatrick et al. 2009). A more recent parameterization of the original model, which can be applied to data with two or more detection methods, might solve some of the identifiability issues (Miller et al. 2011).

15

latter, Sólymos et al. (2012) - of abundance es-timation that avoids the closure assumption and need for multiple surveys at a site by adjusting detection error through covariates that affect detection and occupancy. It performed well against simulated data sets, but remains to be applied in a multispecies framework. Another important advance was the development of a dynamic, multi-season MSOM that models change in occu-pancy between sampling periods as local colonization and extinction parameters, which provides a method to calculate beta diversity and evaluate metacommunity dynamics (Dorazio et al. 2010). When applied to European butterflies, it illustrated the potential of MSOMs to evaluate competing theories of metacommunity ecology (Holyoak et al. 2005). Similarly, development of MSAMs of-fers more than just the ability to improve estimation of abundance-based measures of diversity. They create new opportunities for evaluating the shapes and functional forms of species-abun-dance distributions (Preston 1948; Wilson 1991) and offer applications to niche ecology (King 1964; Whittaker 1965) and neutral theories (Hubbell 2001; Harte 2011).

1.7 Acknowledgements This work was supported by the Berkeley Initiative in Global Change Biology, NSF grants CNH 1115069 and DEB-1051342 to S.R.B. and a David H. Smith Postdoctoral Fellowship to M.W.T. Reviews by Marc Kéry, Paul Craze, Perry de Valpine, the Beissinger laboratory, and three anony-mous reviewers greatly improved this manuscript.

The hierarchical modeling approach of MSOMs, particularly within a Bayesian framework, offers its own challenges. There are few options for choosing among com-peting models that contain different parameterizations. Information-theoretic meth-ods (Burnham and Anderson 2002) and model-selection indices, such as the deviance information criterion (DIC; Spiegelhalter et al. 2002), cannot be reliably calculated for hierarchical models (Celeux et al. 2006; Millar 2009). Although no single model-selec-tion approach for hierarchical models has yet seen wide application (Royle and Dora-zio 2008), it is an area of ongoing development and advancement (e.g., Mattsson et al. 2013). Model selection might present challenges, but variable selection can be done through various methods .

Box 1.3 continued

16

Glossary Bayesian shrinkage: increased precision and accuracy in a parameter estimate that occurs when

parameters are modeled with a common prior distribution, which results in the individual .

False absence: an individual or species that is present at a site but not detected during sampling due either to observation error or to the individual not being available for detection during the survey period.

Hierarchical model: a statistical model described by a nested sequence of observed and unob-served random variables, which is a particularly flexible and transparent way to model com-plex dependencies in observed data.

Hill numbers: a set of mathematically unified diversity indices whose outputs are in units of ef-fective number of species .

Hyperparameter: a parameter that governs the community-level distributions from which spe-cies-specific probabilities are drawn.

Latent variable: a parameter that is inferred from other parameters through statistical models and that is not directly observed by sampling.

Markov chain Monte Carlo (MCMC) methods: a class of algorithms used to generate depend-ent random samples from statistical distributions that may be intractable analytically (typi-cally posterior distributions in Bayesian analysis of a model).

Nonparametric estimator: an estimator that does not assume an underlying distribution for the data on which the estimate is being made.

Rarefaction: a plot of the average number of species encountered as a function of sampling ef-fort, which facilitates comparison of diversity metrics across samples derived from unequal sampling effort.

Species-accumulation curve: a plot that records the total number of species encountered with increasing sampling effort. Effort is usually measured by number of samples or number of individuals.

Uncorrected count: the count of individuals or species from survey data that has not been statis-tically adjusted to account for sampling bias or sources of imperfect detection.

17

2 Collapse of a Desert Bird Community over the Past Century Driven by Climate Change2

Climate change has caused deserts, already defined by climatic extremes, to warm and dry more rapidly than other ecoregions in the contiguous United States over the last 50 years. Desert birds persist near the edge of their physiological limits, and climate change could cause lethal dehydra-tion and hyperthermia, leading to decline or extirpation of some species (Albright et al. 2017). We evaluated how desert birds have responded to climate and habitat change by resurveying historic sites throughout the Mojave Desert that were originally surveyed for avian diversity during the early 20th century by Joseph Grinnell and colleagues. We found strong evidence of an avian com-munity in collapse. Sites lost on average 43% of their species and occupancy probability declined significantly for 39 of 135 breeding birds. The Common Raven was the only native species to sub-stantially increase across survey sites. Climate change, particularly decline in precipitation, was the most important predictor of site-level persistence, while habitat change had a secondary influence. The presence of surface water reduced the loss of site-level richness, creating refugia. Habitat pref-erence and diet were the two most important species traits associated with occupancy change. Car-nivores were more severely impacted than other dietary guilds. Arid-land specialists and habitat generalists were more resilient to decline than grassland and forest species. The collapse of the avian community over the past century may indicate a larger imbalance in the Mojave and provide an early-warning of future ecosystem disintegration, as climate models project an increasingly dry and hot future in this region.

2.1 Introduction Deserts are important bellwethers of climate change. Already defined by climatic extremes, deserts have warmed and dried more rapidly over the last fifty years than other ecoregions, both globally and in the contiguous United States (Figure 2.1 and Figure 2.2; Zhou et al. 2015; Wuebbles et al. 2017). These trends are predicted to continue through the end of the century (Dominguez et al. 2010). Climate change impacts desert species through the direct effects of thermal and hydric stress, and indirectly via impacts on habitat and food resources. Negative effects of increased tem-peratures have been documented for desert birds (du Plessis et al. 2012), mammals (Moses et al. 2012), invertebrates (Crawford 1981), reptiles (Sinervo et al. 2010), and amphibians (Griffis-Kyle 2016). Moreover, because precipitation and primary productivity are strongly linked in deserts,

2 This work has been published previously and is reproduced here with permission from the pub-lisher, the National Academy of Sciences of the United States of America: Iknayan KJ, Beissinger SR. 2018. Collapse of a desert bird community over the past century driven by climate change. Proceedings of the National Academy of Sciences. doi:10.1073/pnas.1805123115.

18

drying mediates productivity declines that can permeate across trophic levels through an entire desert community (Chesson et al. 2004).

Desert birds comprise a species-rich, easily detectable assemblage, and are closely coupled to their physical environment, which makes them suitable indicators of climatic change (Wolf 2000). Although desert birds exhibit some adaptive capacity to tolerate thermal and hydric ex-tremes (Williams and Tieleman 2005), many already persist at the edge of their physiological limits (Albright et al. 2017). Both heat waves and the chronic deleterious effects that high temperatures can have on fitness imperil desert birds (Wolf 2000). Future warming and associated lethal dehy-dration risk alone could extirpate species from the Desert Southwest, particularly small-bodied birds (Albright et al. 2017). Beyond direct physiological effects on desert avifauna, climate change is projected to drive a redistribution of desert habitat (Rehfeldt et al. 2012). Moreover, the projected drying of the Mojave will likely negatively impact prey and plant-food availability for bird species (Grant and Grant 1987; Morrison and Bolger 2002; McCreedy and van Riper 2014; Smit and McKechnie 2015). Studies of the effects of climate change on North American desert birds are

Figure 2.1 Change in Mean Annual Temperature in the Mojave Differences in mean annual temperature between historical climate (1906-1965)and modern (1986-2015) climate averages.

19

limited (Hargrove and Rotenberry 2011a; Hargrove and Rotenberry 2011b; Cruz-McDonnell and Wolf 2016), but continental-scale surveys suggest arid-land birds are in decline (Sauer et al. 2017). However, the drivers of this decline have not yet been evaluated.

We assessed how climate change and other stressors have impacted desert bird populations over the past century by resurveying sites throughout the Mojave Desert that were originally sur-veyed for avian diversity in the early 20th century by Joseph Grinnell and colleagues. About 85% of desert lands in this region are largely undisturbed and ecologically intact (Figure 2.3; Randall et al. 2010), allowing for assessment of the impacts of climate change without the confounding effects of land-use change. Nevertheless, structural changes to habitat induced by grazing and increased severity and frequency of fire are potentially important threats to Mojave birds (Brooks and Pyke 2001). Invasive grasses have been implicated as the main driver underlying increased fire occur-rence in the Mojave (Brooks et al. 2004). However, climate change likely plays an interactive role in altering the fire regime by: (i) lengthening the fire season (Westerling et al. 2006; Weiss et al. 2009), and (ii) promoting spread of invasive grasses through improved growing conditions and

Figure 2.2 Change in Annual Precipitation in the Mojave Differences in annual precipitation between historical climate (1906-1965) andmodern (1986-2015) climate averages.

20

increased number of frost-free days (Weiss and Overpeck 2005). We related changes over the past century in occupancy and richness of the Mojave avian community to changes in climate (annual precipitation, average temperature, and maximum temperature), climatically-derived measures of ecosystem productivity facilitated by actual evapotranspiration (AET) and inhibited by climate water deficit (CWD), disturbance (fire and grazing), and recent weather (precipitation, average temperature, and maximum temperature of the previous year) using a dynamic multispecies oc-cupancy model (Dorazio et al. 2010). The analysis allowed us to quantify species-level and com-munity change simultaneously, while accounting for species and survey differences in detectabil-ity.

For breeding birds of the Mojave, we examined (i) the relative importance of climate change, ecosystem productivity, habitat disturbance, and recent weather conditions to avifauna

Figure 2.3 Habitat Quality in the Mojave Desert Land in the Mojave Desert characterized by conservation value. Ecologically core and ecologically intact lands are functionally equivalent and both largely undisturbed and unfragmented, however, core habitat is the land considered critically important for the protection of Mojave biodiversity. Moderately degraded land has been impacted by human activities (e.g., off-road vehicle use) but provides some ecological function and habitat for native species. Highly converted land has been dramatically altered from original conditions (e.g., from urban and agricultural development). Data:The Nature Conservancy (Randall et al. 2010)

21

change; (ii ted for change; and (iii) the persistent characteristics of sites that maintain diversity. We predicted that persistence of the avian commu-nity would be negatively related to recent and long-term increases in temperature and disturbance, and positively related to increases in precipitation and habitat productivity. We also predicted spe-cies would experience less decline in occupancy than their counterparts if they had higher mobility (nonterritorial), higher dispersal capacity (migratory), greater plasticity in resource use (widely-distributed species and habitat/diet generalists), greater reproductive potential, and were better adapted to physiological stress (large-bodied) and to desert environments (low-elevation species and arid-land specialists; (Angert et al. 2011). Additionally, we predicted that sites at higher eleva-tions and latitudes, and those with surface water would support higher levels of avian diversity in the modern surveys. Finally, we examined whether the fate of sites varied by management unit to better inform conservation needs. We predicted that sites managed by the US National Park Ser-

Figure 2.4 Study Area and Survey Locations. Federal lands outlined: National Park Service (NPS), National Forest Service (NFS), Fish and Wildlife Service (FWS), and Bureau of Land Management (BLM); with lands managed by NPS labeled: Death Valley National Park (DEVA), MojaveNational Preserve (MOJA), and Joshua Tree National Park (JOTR).

22

vice (NPS) would fare better than sites on multiuse public lands [i.e., US Bureau of Land Manage-ment (BLM) and US Forest Service (USFS)] because of the greater level of protection that park status bestows.

2.2 Methods

2.2.1 Historic Surveys

Surveys of avian diversity were performed at 61 sites across the Mojave Desert spanning a broad latitudinal (33.6° - 37.5°), longitudinal (-117.7° -114.8°), and elevational range (-17.8 3396.2 m). The majority of sites (58 of 61) were on federal lands in Death Valley and Joshua Tree National Parks, Mojave National Preserve, BLM, USFS, and US Fish and Wildlife Service (Figure 2.4). Only one site experienced development from the historic time period; resurvey of this site occurred < 1 km away in habitat matching historic descriptions.

Historic surveys were conducted between 1908 and 1968, with 70% of the surveys occurring between 1917 and 1947. Historic survey data (species encounters, geographic locations, and

of Vertebrate Zoology (available online). Photographs, maps, and site descriptions within note-books allowed identification of historic survey locations with a high degree of accuracy. For exam-ple, during the resurvey of the Yosemite transect, the mean estimated spatial error for locating the true historic sites was estimated to be 0.7 km (Wieczorek et al. 2004).

Historic surveys followed a precursor of the line-transect method and provided a record of detection/nondetection data for all species encountered. Historic data at each site were restricted to surveys that occurred within the breeding season of a single year to ensure closure during the primary sampling period. To reduce the bias that a small number of heavily-sampled sites might have on the estimation of detection probabilities, we limited historic data to a maximum of 10 visits within a single year. Seven historic surveys followed a removal protocol (MacKenzie et al. 2006), and were included in the analysis as such.

2.2.2 Modern Survey Methods

Avian resurvey sites follow the standard sampling protocol used in the Grinnell Resurvey Project (Tingley et al. 2012; Tingley and Beissinger 2013), consisting of 2.25 km transects, each with 10 point count stations located 250 m apart. Systematic bird inventories were conducted using varia-ble-distance circular plot, seven-minute point counts, where all species detected by sight and sound are recorded (Ralph et al. 1995). Fifty-five sites were selected from the historic data based the presences of repeated survey effort for avian diversity. To minimize bias between the historic and modern survey data, characteristics of the historic surveys were matched when resurveying mod-ern sites: geographical location and extent of survey sites, elevational range covered, habitats sur-veyed within sites, and timing of the survey during the breeding season. While fine-scale cover data was not available for the historic surveys, vegetation at the modern survey sites qualitatively

23

matched the historic descriptions of habitat and plant species present. Therefore, the effects on detectability due to vegetation structure is expected to be similar between time periods. Survey transects were visited three times within a year allowing for the calculation of detection probability separate from the probability of occurrence. We conducted modern surveys in the same season as historic surveys. However, we were unable to exactly match the date of surveys, so Julian day and its quadratic were included as a detection covariate to enable detectability to vary within the breed-ing season. Inclusion of the quadratic term allowed detectability to reach a peak during the breed-ing season, in line with known behavior of birds (Furnas and McGrann 2018). Variation in detect-ability unaccounted for by survey-level covariates and characteristic matching was captured through inclusion of a categorical covariate for time period in the multispecies occupancy model.

The historic data set encompassed 2371 species encounters during 204 visits at 55 sites, with a mean of 3.46 visits per site. The modern data included 2661 species encounters during 183 visits to 61 sites, with 3 visits per site. A total of 135 breeding species were included in the analysis. Potential for breeding was determined by overlap of expert-defined breeding ranges (Rodewald 2015) with survey sites. Nocturnal birds, obligate waterbirds, and nonbreeding birds were not ad-equately surveyed by our methods and were excluded from analysis. Species were analyzed using a revised taxonomic treatment. Museum specimens provided a verifiable record of a species oc-currence at a particular locality, and allowed accurate translation of historic taxonomic nomencla-ture into current species definitions. Reclassification of the recently- s Sparrow and Sagebrush Sparrow was not unequivocal, thus were aggregated in the analysis as the historic species complex the Sage Sparrow. Four species the nonnative European Starling, Chukar, and Eurasian Collared-Dove, and the invasive Great-tailed Grackle known to colonize the study area after the completion of historic surveys (Small 1994) were treated as being truly absent from the historic data.

2.2.3 Occupancy Model for Community Dynamics

A dynamic, multispecies, occupancy model without augmentation (Dorazio et al. 2010) was used to explicitly model the colonization and persistence dynamics that occurred between the historic and modern time periods. Survey data, denoted as ( = 1 for a detection and = 0 for a nondetection of a species ( ), at site ( ), time period ( ), and survey ( ), were assumed to be the result of the imperfect observation of the true incidence of a species at a site during a time period:

( = 1 if present, and = 0 if absent). Given the detection probability , the observation model is:

(2.1)

where is always zero if the species is absent from the site (i.e., = 0), but otherwise the species is detected with the probability . The pattern of detections and nondetections of a species across repeated surveys at a site within a time period, assuming closure of that site to immigrations

24

model. The initial incidence of a species at a site is modeled as a Bernoulli process governed by that

):

(2.2)

Ecological dynamics that occurred between the historic and modern time periods were modeled as a first-order Markovian process (i.e., the incidence of a species in the modern period was as-sumed to depend on the incidence state of the historic period), which can be formalized as a func-

) and colonization ( ) probabilities:

(2.3)

More simply, if the site was occupied ( = 1) in the historic period, the only component governing in the

formula becomes zero. Conversely, if the site was unoccupied ( = 0), the only component govern-

considered one transition period, however, this model form is generalizable to additional time steps.

Covariates were included as a linear combination of effects with a logit-link transformation for each of the four probabilities estimated by the model [detection ( ), historic occupancy ( ), local persistence ( ), and local colonization ( )]:

(2.4)

(2.5)

(2.6)

(2.7)

where the naught terms in each equation represent the species-specific intercept, while the bold Greek terms represent arrays of coefficients and their associated covariates represented by the bold Latin terms. The species-specific intercept and coefficient values were governed by normal distri-butions, which were in turn drawn from community-level hyperdistributions. For each of the com-munity-level intercept and coefficient priors, the mean was an assigned a vague priors:

(2.3)

We fit the model with a weakly-informative prior applied to the variance of the coefficient hyperdistributions:

25

(2.3)

which behaves as a type of regularization also known as shrinkage (Hooten and Hobbs 2015).Reg-ularization limits the effects of collinearity in the covariates and prevents over-fitting, allowing us to evaluate a global model of community response (Hooten and Hobbs 2015). For full model and prior specification, see Appendix A.

2.2.4 Model Covariates

To evaluate community dynamics between time periods, we fit covariates to each probability mod-eled. Survey era was included as a categorical covariate of detection to explicitly capture differences between the two time periods, rather than having the species-specific intercepts and covariates implicitly vary by time period (Dorazio et al. 2010). Additionally, Julian day and its quadratic term were included in the detection model to allow for detectability to vary during the breeding season. Initial occupancy covariates included elevation and its quadratic term, and historic climate aver-ages (annual precipitation, mean annual temperature, and maximum annual temperature). Since we rarely observed colonization and the model-estimated mean colonization probability was very low ( = 0.003; 95% CRI: <0.001 - 0.009), we only included one nuisance covariate on colonization to avoid overfitting - years elapsed between historic and modern surveys. This covariate was also included in the model for persistence.

Covariates chosen to assess the drivers of community persistence fell into three categories: climatic change (annual precipitation change, annual mean temperature change, annual maximum temperature change), recent weather (precipitation, mean temperature and maximum tempera-ture of the year prior to the modern surveys), and habitat change (grazing intensity, mean fire return interval, climate water deficit change, actual evapotranspiration change). Vegetation de-scriptions from the historic surveys were qualitative in nature, so we were unable to evaluate effects of fine-scale habitat change on community dynamics. Environmental covariates were calculated from the Basin Characterization Model (Flint et al. 2013), using a 5 km window to capture local variability. Change covariates were calculated from differences between historical climate (1906-1965) and modern (1986-2015) climate averages. Historic climate averages and weather of the pre-vious year were also extracted from the BCM for relevant years. Mean fire return interval for each

Program (http://www.frap.cdf.ca.gov) and the Nevada BLM. Current grazing intensity data (ani-mal unit month per hectare, AUM) were assembled from information provided by NPS, USFS, and BLM regional offices. All covariates were centered and scaled prior to inclusion in the model.

Model data were preprocessed in R version v3.4.0 (R Core Team 2017) and fit using MCMC as implemented by JAGS v4.2.0 (Plummer 2003). Forty parallel chains were run for 32,500 itera-tions, retaining every 100th iteration, resulting in 13,000 posterior samples. The model was fully adapted (~100 iterations) and burn-in was run for 10,000 iterations for each chain. Convergence was assessed by visual inspection of trace plots and by using the Gelman-Rubin statistic ( ); values

26

also known as the Bayesian p-value) to test goodness-of-effects that have a high probability of being nonzero (i.e., 95% CRIs nonoverlapping with zero), or to describe two values that have a high probability of being different (i.e., nonoverlapping 95% CRIs). See Appendix A for PPC calculation details.

2.2.5 Species Traits as Predictors of Change

Species-specific traits hypothesized to affect response to climate change were evaluated for their potential to account for occupancy change: dietary breadth, fecundity, elevational preference, range margin, dispersal ability, body size, and territoriality (Angert et al. 2011). Traits were com-piled from the following sources (Table A.1): migratory mode, elevational preference, range mar-

The Birds of North America Online (Rodewald 2015); habitat specialization from Partners in Flight (2017); diet breadth from De Graaf et al. (1985); and clutch size and body mass from Myhrvold et al. (2015). Range margins were categorized by the presence of a range boundary within the study area, as defined by expert-created distribution maps (Rodewald 2015). Species with restricted ranges were considered to have

ors of the relative change in occupancy ( between historic and modern surveys using weighted linear regressions. To account for uncertainty in occupancy estimates, weights were cal-culated from the inverse of the variance of the posterior distribution of , giving higher weight to estimates with lower uncertainty (Hedges et al. 1999). We transformed the data [ ]) to linearize the ratio for both increases and decreases in occupancy (Hedges et al. 1999) and added 0.016, the equivalent of one site being occupied, to the historic and modern occupancy probability of each species to adjust for zero occupancy estimates that occurred for some species in either time period.

Phylogenetic signal was evaluated to ascertain the independence of species in the model and to identify taxonomic clades that experienced similar responses. We downloaded 500 trees from birdtree.org, 250 of each of the Ericson and the Hackett backbones (Jetz et al. 2014). We

gnal in relative occupancy change was present across the entire evolutionary tree and reported the median p-value for all 500 trees (Blomberg et al. 2003; Frishkoff et al. 2014). We also examined whether response was clustered within taxonomic families, as defined by the American Ornithological Society, through a random-ization test; was calculated for each family and compared to 10,000 random draws of the same number of species as in the family. A family was considered significantly different when was greater than or less than 95% of the random draws (Frishkoff et al. 2014). Nonnative species intro-duced after the historic surveys were excluded from the from the phylogenetic and traits analyses because their change in occupancy was assumed to be related to invasion biology.

27

2.2.6 Diversity and site characteristics

Species richness was calculated as the sum of the species incident at the site during each time pe-riod. Site-level characteristics (elevation, latitude, management unit, and presence of surface wa-ter) were tested as predictors of relative change in richness ( ) using weighted linear regressions in the same fashion as .

2.3 Results

2.3.1 Change in the Mojave Avian Community over the Past Century

Change in the Mojave avian community over the past century was characterized by species loss without replacement at the resurveyed sites (Figure 2.5). A strong signal of avifaunal decline is apparent from the community-level measures estimated by the dynamic multispecies occupancy model, which was a sufficient fit to the data (PPC = 0.13). Across the community, the mean prob-ability of species persistence at occupied sites between the two time periods was moderately low ( = 0.43; 95% CRI: 0.31 0.55), and the mean probability of species colonizing unoccupied sites was close to zero ( = 0.003; 95% CRI: <0.001 - 0.009). Only the Common Raven and Wren had colonization probabilities significantly greater than 0.10. On the other hand, species-level persistence probabilities at sites were uniformly distributed across the community with spe-cies mean values ranging from 0.03 0.96. See Table 2.1 and Table A.2 for the summary of the posterior distributions for the community- and species-level hyperparameters.

Figure 2.5 Change in Occupancy and Richness in Mojave Avifauna over the Past Century. Points are posterior distribution means: blue represents significant increase [non-overlapping historic and modern 95% credible intervals (CRI)], red represents significant decreases, and orange represents species truly absent from the historic surveys. (A) Historic and modern occupancy probabilities for 135 bird species. Error bars were omitted for visual clarity. (B) Historic and modern site richness for 61 resurvey sites. Error bars represent 95% CRI.

28

The interplay of colonization and persistence dynamics resulted in an average modern oc-cupancy probability ( = 0.11; 95% CRI: 0.07 0.15) that was significantly lower than historic occupancy ( = 0.24; 95% CRI: 0.17 - 0.32). Of 135 species, occupancy declined significantly for 39 and increased significantly for only one, the Common Raven (Figure 2.5A). Four species absent historically the nonnative European Starling, Chukar, and Eurasian Collared-Dove, and the in-vasive Great-tailed Grackle now occur at a limited number of sites (mean : 0.02 - 0.15). Species with large significant decreases in occupancy probabilities (>0.30) included American Kestrel, Northern Mockingbird, Prairie Falcon, Turkey Vulture, Western Wood-Pewee, White-throated Swift, Western Kingbird, Chipping Sparrow, Mourning Dove, Sharp-shinned Hawk, Violet-green Swallow, and Brewer's Sparrow. Most were widespread historically ( >0.50). Forty additional species (30%) experienced nonsignificant declines in mean occupancy probability by 0.20-0.35 (n = 18) or by 0.10-0.20 (n = 22).

Occupancy declines translated into site-level losses of species richness (Figure 2.5B). Sites decreased on average by 17.9 bird species (95% CRI: -20.5 -15.5), or 42% of their richness, from the historic time period. As a result, 55 of 61 sites experienced a significant decline in richness. This drop in species richness was the result of a mean loss of 20.9 bird species per site (95% CRI: -13.0 -29.3) and a mean gain of 3.0 bird species (95% CRI: 0.6 5.6).

The decline in avian occupancy and richness from the historic to the modern era was not due to lower detectability of species during our resurveys. Species present at a site during modern surveys were very likely to be encountered at least once ( = 0.94; 95% CRI: 0.91 - 0.96); this was a significantly higher level of site-level detection than for historic surveys ( = 0.57; 95% CRI: 0.48 - 0.65).

2.3.2 Drivers of Mojave Avifauna Change

Decline of Mojave birds over the past century was driven primarily by climate change and second-arily by measures of ecosystem productivity (Table 2.2, Figure 2.6). A long-term reduction in pre-cipitation best described changes in site-level persistence of Mojave birds and significantly ac-counted for persistence for 35 species (Table 2.2). Maximum temperature and precipitation in the year prior to resurveys affected community persistence to a lesser degree, describing significant

Table 2.1 Posteriors of Community Probabilities Summary of posterior distribution of community-level hyperpa-rameters in dynamic multispecies occupancy model.

Intercept mean Parameter Mean 95% CRI Mean 95% CRI Detection 0.60 0.55, 0.66 0.96 0.19, 0.64 Occupancy 0.24 0.17, 0.32 2.12 -1.56, -0.75 Persistence 0.43 0.31, 0.55 1.92 -0.79, 0.21 Colonization 0.00 0, 0.01 2.62 -7.23, -4.68

29

declines of only five and two species, respectively. Moreover, contrary to expectations, community persistence was positively related to higher maximum temperatures in the year prior to resurveys. Long-term climatic change of mean and maximum temperatures, and mean temperature in the year prior to resurveys were not important predictors of community persistence. Only one habitat factor was related to community persistence change in AET. Persistence of 14 species responded negatively to change in AET (Table 2.2, Figure 2.6).

Declines in avian occupancy were distributed relatively evenly throughout the avian evo-lutionary tree (B p = 0.45). However, a randomization test found variation in occu-pancy change clustered in a few families (Figure 2.7). Corvidae, Troglodytidae, and Polioptilidae

Figure 2.6 Drivers of Mojave Avian Community Persistence The relationship of climate change, recent weather, and habitat change to community-levelprobability of persistence. Shaded area represents 95% credible intervals (CRI). Asterisked panels represent covariates whose 95% CRI do not overlap zero.

30

declined significantly less than other families, while Accipitridae and Falconidae experience sig-nificantly more decline.

Several species traits were related to occupancy declines (Table 2.3, Figure 2.8). The two most important predictors of occupancy change were habitat and diet specialization (Table 2.3). Together they accounted for a total AIC weight of 0.89. Arid-land specialists and habitat generalists declined significantly less than species inhabiting grasslands and forests (Figure 2.8A). Carnivores declined significantly more than insectivores and herbivores, while omnivores declined signifi-cantly more than herbivores (Figure 2.8B). There was less support for effects of range margin: spe-cies with patchy distributions declined significantly more than species without a distributional limit in the study area (Figure 2.8C). Migratory status, elevational preference, territory type, body mass, and clutch size were poor predictors of occupancy decline (Figure 2.8D-H).

Two key site-level characteristics were associated with reduced rates of species loss for Mo-jave birds (Table 2.4, Figure 2.9). The presence of surface water was the most important predictor of relative change in richness (Table 2.4), and sites with surface water present experienced 13.4% less decline than sites without surface water (Figure 2.9A). Management unit was next most im-portant factor, but protected parks lands were not necessarily less immune to species loss than other public lands. Sites in Death Valley National Park incurred the greatest declines in relative richness compared to all other management units, which did not differ (Figure 2.9B). Higher levels

Table 2.2 Relationship of Avian Persistence to Covariates Coefficients of community-level persistence probability from a dynamic multispecies occupancy model. Coefficients whose 95% credible interval do not cross zero, are bolded. The count of species with coefficients not overlapping zero are summarized by the direction of the response.

Species

Responses Covariate 95% CRI (-) (+) Climate Change 1.12 0.43, 1.86 0 35 0.30 -0.01, 0.65 0 0 -0.04 -0.37, 0.32 0 1 Weather in Previous Year Precipitation (mm) 0.62 0.19, 1.08 0 2 Mean temperature (°C) -1.44 -3.17, 0.19 6 0 Max temperature (°C) 1.71 0.01, 3.53 0 5 Habitat Change Grazing intensity (AMU) 0.17 -0.15, 0.63 0 0 Fire return interval (years) -0.16 -0.47, 0.12 0 0 0.30 -0.37, 0.90 0 0 AET (mm) -0.86 -1.66, -0.12 14 0

31

of richness were maintained somewhat better at higher elevation sites, with 6.3% less decline in richness for every 1000 m in elevation gain (Figure 2.9C). Latitude had little influence on relative richness loss over the geographic scale we investigated (Figure 2.9D).

2.4 Discussion

2.4.1 Collapse of a Desert Bird Community

Birds of the Mojave Desert exhibited a dominant signal of occupancy decline and erosion of site-level species richness over the last century (Figure 2.5). Both occupancy probability and site-level

Figure 2.7 Relative Occupancy Change by Taxonomic Family Phylogenetic tree of species in analysis. Families arranged in evolutionary order, colored by family. Blue lines indicatefamilies with significantly less decline than other families. Red lines indicate families with significantly more decline than other families

32

diversity diminished by about half: the average species occupancy probability declined from 0.24 to 0.11, and sites lost on average 43% of their species. The decline of birds in the Mojave Desert was exceptional relative to changes observed over the same timescale for resurveys conducted in other regions of California following the same methods. Across the Sierra Nevada (Tingley and Beissinger 2013), nearly as many sites gained (43%) as lost bird species (57%), resulting in a small decline in average site-level richness (-2.5 species per site) compared to the Mojave (-18.4 species per site). In resurveys of the Central Valley, richness was stable with more bird species significantly increasing (n = 35) than decreasing (n = 27) in occupancy (MacLean et al. In review). In contrast, there was no counterbalance to decline of species in the Mojave, where only the Common Raven increased significantly in occupancy. The trends we observed in the Mojave are corroborated by findings of the North American Breeding Bird Survey (BBS). There has been a 46% decline in the abundance of arid-land indicator species since the onset of BBS surveys in 1968 (NABCI 2014; Sauer et al. 2017).

The nearly uniform decrease in site-level richness across the biome and the dominant sig-nal of decline in the occupancy of many species signifies the potential emergence of a new, lower carrying capacity for birds in the Mojave Desert ecosystem. Taken together, these results suggest a community in the process of collapse. Community or ecosystem collapse is usually the result of multiple causes that create synergistic effects and feedbacks (Cardinale et al. 2012). For example, climate change acted in concert with an emerging disease to threaten global amphibian communi-ties (Price et al. 2014), with herbivory to threaten Arctic avifauna (Ims and Henden 2012), and with a complicated suite of threats to imperil Hawaiian avifauna (Paxton et al. 2016). Nevertheless, climate change alone can mediate the collapse of communities (Bergstrom et al. 2015). Whether the collapse of Mojave birds will continue is unknown, although declines of arid-land birds in BBS

Table 2.3 Species Traits and Occupancy Declines Weighted linear regression models examining the relationship of species-specific traits to relative occupancy change in 135 bird species in the Mojave Desert, over the last century. An intercept-only model was also evaluated and is represented as (.).

Parameter log ℒ K AICc wi Habitat -106.7 5 224 0.0 0.646 Diet -107.7 5 225.9 1.9 0.248 Range margin -108.7 5 227.9 3.9 0.091 Elev. pref. -112.9 3 232.1 8.1 0.011 Migratory -113.3 4 234.8 10.9 0.003 Territoriality -116.5 3 239.2 15.3 0.000 (.) -118.2 2 240.5 16.6 0.000 Body mass -117.4 3 241 17.1 0.000 Clutch size -118.0 3 242.2 18.2 0.000

33

may have abated over the past decade (Sauer et al. 2017). Nevertheless, populations have not re-covered to initial levels, nor have they approached the historic baselines that our surveys revealed.

2.4.2 The Drivers of Collapse

Climate change, particularly decline in rainfall, was the most important driver of avian community dynamics in the Mojave Desert. Sites that received less precipitation in recent decades compared to early in the 20th century had higher local extinction probabilities (Figure 2.6). Most sites became drier ( = -3.6%, = 5.9%; Figure 2.2), receiving up to 20% less precipitation. The causal relation-ship between precipitation and avian population dynamics has many linkages, given the wide-spread effects precipitation has on arid communities (Chesson et al. 2004). Dry periods can cause desert birds to operate at water deficits due to decreased access to water-rich food resources, which can negatively impact fitness (Smit and McKechnie 2015). Reproduction and survival in many arid-land birds are positively related to precipitation, a response likely mediated through effects on food availability, predation, and nest parasitism in arid environments (Grant and Grant 1987; Morrison and Bolger 2002; McCreedy and van Riper 2014). Rainfall in the year prior to resurveys also had a positive relationship with local persistence, but the effect was weaker than the influence of long-term change in precipitation (Table 2.2). Even though California as a whole experienced a historic drought from 2012-2017, the most severe impacts occurred outside of the Mojave Desert (Williams et al. 2015), and in the year prior to our resurveys rainfall was within normal variation (i.e., within 1 of 30-year averages) for all but three of our sites.

Figure 2.8 Relationship between Species' Traits and Relative Occupancy Change Panels are ordered according to AICc weights in a single-covariate model set (Table 2.3) for: (A) habitat specialization; (B) diet specialization (carnivores, omnivores, insectivores, and herbivores); (C) range margin based on distributional extent in the study area; (D) high elevation species versus those that reside on the desert floor; (E) migratory habits; (F) territoriality; (G) body mass; and (H) clutch size. Error bars represent 95% confidence intervals around estimated means.

34

Climate change influenced the collapse of the Mojave bird community primarily through precipitation change rather than temperature change. Long-term warming was not strongly related to avian decline, but there was an influence of maximum temperature in the year previous to re-surveys (Table 2.2). Contrary to our expectations, however, community persistence was positively related to the maximum temperature in the year prior to resurveys, which could occur for several reasons. First, due to topographic complexity in the Mojave, there was high variability in site-level maximum temperatures in the year before resurveys ( = 31.8°C, = 5.6°C). Warming at the higher elevation sites, may translate to a longer breeding season leading to greater reproductive success which may positively affect the community (Pearce-Higgins et al. 2015). Second, observed temperatures may not yet be hot enough to have the negative impacts we anticipated. The physio-logical costs of heat avoidance and dissipation behaviors begin to negatively impact fitness of some desert birds when temperature exceeds 35°C (Albright et al. 2017; Pattinson and Smit 2017); at temperatures > 40°C, evaporative water loss accelerates and puts birds at risk of lethal dehydration (du Plessis et al. 2012; Cunningham et al. 2015; Smit et al. 2016). These thermal limits were ex-ceeded at only 26% and 9% of our sites, respectively. Albright et al. (2017) suggested that body size would be an import predictor of decline and extirpation in desert bird communities based on its relationship with heat-related dehydration. Body size, however, was a poor predictor of decline in our study (Table 2.3), which also supports our conclusion that the thermal environment did not appear to be a substantial contributor to Mojave bird declines.

Habitat disturbance and measures of ecosystem productivity were less important than cli-mate change in driving avian community collapse. Although fire and grazing can impact birds in arid systems (Fleishman et al. 2006; Esque et al. 2013), neither strongly influenced avian commu-nity response in our study. Only change in actual evapotranspiration (AET) was related to avian persistence and, contrary to our expectations, it had a negative relationship at the community level and for 14 species (Figure 2.6 and Table 2.2). AET is calculated using a combination of climatic, topographic, and edaphic conditions related to plant growth (Rapacciuolo et al. 2014), and is strongly correlated with annual net primary productivity in the Mojave Desert (Lane et al. 1984). Our results imply that some Mojave birds responded negatively to long-term habitat change in the

Table 2.4 Site Characteristics and Loss of Richness Weighted linear regression models examining the relationship of species-spe-cific traits to relative occupancy change in 135 bird species in the Mojave De-sert, over the last century. An intercept-only model was also evaluated and is represented as (.).

Parameter log ℒ K AICc wi Surface water 10.8 3 -15.3 0.0 0.577 Management 12.4 5 -13.6 1.6 0.257 Elevation 8.8 3 -11.1 4.2 0.071 (.) 7.6 2 -11.0 4.2 0.069 Latitude 7.7 3 -9.0 6.3 0.025

35

form of vegetation growth. Precipitation is used in calculating AET, and the opposing direction of the community response to these two factors appears conflicting. However, species that responded negatively to a change in AET differed from those species that responded positively to precipitation change. Heterogeneity among species in response to environmental change is relatively common in avian communities (Huang et al. 2017), and is expected given the taxonomic diversity that com-prises the Mojave bird community.

2.4.3 Winners and losers

There was an apparent lack of climatological among the Mojave avifauna. The only species occurring in historic surveys that increased significantly was the Common Raven, likely due to nonclimatic causes (e.g., the proliferation of anthropogenic food resources in the desert; 44). Relative change in occupancy was randomly dispersed across the phylogenetic tree, with corvids, wrens, and gnatcatchers declining less severely than others (Figure 2.7). Arid-land spe-cialists, habitat generalists, and more broadly distributed species were also more resilient to de-cline, but these groups still decreased, just to a lesser degree (Figure 2.8). The void left by declining species in other regions affected by climate change has been filled, at least partly, by an influx of species for whom the region has become more favorable (Tingley and Beissinger 2013; MacLean et al. In review). While the Sonoran Desert immediately south of the Mojave holds the anticipated pool of more desert-hardy species, we found no expansion or increase of Sonoran-representative birds already present in the study area (e.g., Phainopepla, Verdin, and Black-throated Sparrow). Nonnative and invasive species colonizing the Mojave in the late 20th century experienced only limited occupancy gains (Figure 2.5).

Declines in occupancy occurred broadly in Mojave birds, including species with a wide range of historic occupancies and evolutionary histories (Figure 2.5). Carnivores were impacted more severely than other dietary guilds (Figure 2.8). This trend was also reflected in our family clustering analysis, in which falcons and accipiters experienced significantly greater declines than

Figure 2.9 Relationship between Site Characteristics and Relative Richness Change -covariate analysis (Table 2.4): (A) the pres-

ence of surface water; (B) land management unit [Death Valley National Park (DEVA), Joshua Tree National Park(JOTR), and Mojave National Preserve (MOJA)]; (C) elevation in meters (D) latitude in degrees. Error bars and shaded area represents 95% confidence intervals around estimated means.

36

other families (Figure 2.7). This result is in line with theoretical and empirical evidence of greater sensitivity of higher trophic levels to climate change, either due to amplification of declines at lower trophic levels or to a temporal mismatch with prey resources (Berggren et al. 2009). Loss of pred-ators has led to trophic downgrading and greatly altered the structure and function in other eco-systems (Cardinale et al. 2012).

2.4.4 Conservation Implications

Our results provide strong evidence that bird communities at sites in the Mojave Desert are col-lapsing to a new, lower baseline that supports about half the species richness that was present a century ago. While our study only captures the decline of a single taxon, the disintegration of avian biodiversity may be an indication of a larger imbalance in the Mojave and an early-warning indi-cator of future ecosystem collapse. Birds serve many functional roles in ecosystems (Whelan et al. 2008), and biodiversity loss can have wide-ranging effects (Cardinale et al. 2012). Collapses can trigger feedbacks that cause secondary extirpations, and can lead to a transition to a new ecosystem state (Cardinale et al. 2012). Moreover, the warming and drying that is unequivocally projected to occur over the next century in the Mojave Desert could further exacerbate or compound the cur-rent decline in the avian community (Dominguez et al. 2010; Albright et al. 2017). Protecting sites with higher levels of precipitation, with surface water, and at higher elevations (Table 2.4 and Fig-ure 2.9) may provide refugia in the face of future climate change. Moreover, desert springs across the region are drying from continued groundwater removal (Patten et al. 2008). Changes in water management policies are needed to reverse this trend. Installation of artificial water sources may help mitigate hydric stress and buoy avian diversity (Cutler and Morrison 1998), but it is a tempo-rary measure.

The collapse of the Mojave bird community has occurred without regard to the protected status of desert lands. The strongest declines occurred in one of the largest National Parks, Death Valley, where 90% of the land is designated as wilderness. Protected areas are only partly effective in protecting habitat (Easterday et al. 2018) and mitigating population declines caused by the broad-reaching consequences of human activities (Barnes et al. 2016), and combating climate change will require global policy efforts in addition to local actions to reduce greenhouse gas emis-sions. Although the Mojave is still relatively undeveloped, it is currently experiencing considerable pressure from the renewable energy sector (Cameron et al. 2012). Balancing protection of biodi-versity and reducing dependence on fossil fuels will require smart development of green energy in the Mojave.

2.4.5 Acknowledgements

This research was performed with support from the NSF (DEB-1457742, DEB-1501757, DGE-1106400), National Geographic Society, NPS, Museum of Vertebrate Zoology and Dept. of Envi-ronmental Science, Policy and Management at UC, Berkeley. B.O. Wolf, P. Unitt, L. Hargrove, and

37

M.W. Tingley provided valuable comments on the manuscript. L. J. Barth, T. S. Baxter, R. A. Klotz, and C. N. Hawk assisted with bird surveys.

38

3 Making the Transition: Avian Biogeographic Response to Climate Change across Biomes

but ecology as a whole: the idiosyncrasies of ecological landscapes can mean findings from one system may not be relevant to any other. Cross-system comparisons can verify that trends are more than just regional in nature, which legitimizes their application to the development of broad-scale predictive models or management recommendations. We evaluated cross-system community dy-namics by revisiting sites that span two biomes and the transition zone between them that were originally surveyed for avian diversity in the early 20th century. The Mojave, a warm desert, and the Great Basin, a cold desert, have distinct biotic assemblages and meet along a wide, roughly east-west boundary. The contiguity of the deserts creates an ideal place to study how climate change is differentially impacting communities across biomes. We sought to understand distributional change at the Mojave-Great Basin transition zone, and to evaluate whether species of the warm deserts, which continues to warm, are expanding into the cold desert with less warming, or whether the transition zone behaves as a barrier to northward expansion. We revisited 106 sites across both deserts and quantified change for 162 breeding bird species using a dynamic multispecies occu-pancy model. Individual species experienced occupancy changes in consistent directions in both deserts. A substantial proportion of species were in decline in one or both deserts (39.5%), while relatively few species were increasing (6.2%). Most species (n = 80) shifted one or both of their latitudinal range limits. Range shifts were highly idiosyncratic in nature, causing the avifaunas of the two deserts to be less strongly structured than they were in the past. The redistribution of spe-cies is driving the genesis of novel communities which may have ecological consequences that are yet to be realized.

3.1 Introduction Anthropogenic climate change has resulted in both poleward and upward shifts in elevation (Chen et al. 2011), but responses have been heterogeneous. Some species show no range response, while others have shifted in directions opposing expectations based on climate warming alone (Tingley et al. 2012; Rapacciuolo et al. 2014). Climatic factors beyond temperature (Tingley et al. 2009) and nonclimatic factors, such as habitat change (Preston et al. 2008), species traits (Chen et al. 2011), and biotic interactions (Post 2013) complicate patterns of range shift. Conservation strategies re-sponding to climate change would benefit from taking a community-level rather than the single-species approach that has been the dominant mode of prediction (Walther 2010; Mokany and Fer-rier 2011). Moreover, becauconsequences of climate change (Levin 1992), the idiosyncrasies of ecological landscapes can mean findings from one system may not be relevant to another. Cross-system comparisons can verify that trends are more than regional in nature, which legitimizes their application to the develop-ment of broad-scale predictive models or management recommendations.

39

We evaluated cross-system community dynamics by revisiting sites that span two biomes and the transition zone between them that were originally surveyed for avian diversity in the early 20th century. The Mojave, a warm desert, and the Great Basin, a cold desert, meet along a sinuous, roughly east-west boundary approximately 450 km wide stretching from southern California

Figure 3.1 Study Area and Survey Locations. Federal lands outlined: National Park Service (NPS), National Forest Service (NFS), Fish and Wild-life Service (FWS), and Bureau of Land Management (BLM).

40

across Nevada to Utah (Figure 3.1). Compared to the Great Basin, the Mojave has lower desert-floor elevations, hotter summers, more moderate winters, and receives most of its precipitation in the form of winter rainfahot but less extreme summers, colder winters, and receives most of its precipitation via snow (MacMahon 1979; Brussard et al. 1998; Peel et al. 2007). The boundary between the Great Basin and Mojave is considered the most distinctly delineated boundary in the North American deserts: topographically, climatologically, and biotically (MacMahon 1979). Biogeographically, it is zone where coincident turnover of warm- to cold-desert species has been noted across taxa: plants (Beat-ley 1975), mammals (Kelt 1999), birds (Behle 1978), insects (Tinkham 1948), and reptiles and am-phibians (MacMahon 1979). Even though the bird communities of the Great Basin and hot deserts have little endemism, they have long been recognized as distinct sets of avifauna with abrupt tran-sitions between them (Miller 1951; Udvardy 1963; Behle 1978). The contiguity of these deserts creates an ideal place to study if 20th century climate change is differentially impacting communi-ties across biomes.

Climate change has already impacted both deserts, and its effects are projected to amplify in the future. Over the past century, the Mojave has experience a ~2°C increase in mean tempera-tures, while the Great Basin has also warmed but to a lesser extent (1.0°C increase (Bai et al. 2011; Tang and Arnone 2013). There is a great amount of spatial variability in precipitation in both de-serts, but overall the Mojave has dried (Bai et al. 2011) while the Great Basin has become wetter (Xue et al. 2017). The Southwest as a whole is anticipated to experience substantial warming during the next century (+2.5-3°C), with more warming occurring in the Great Basin than elsewhere. Continued drying is projected for the Mojave, with the northern portions of the Great Basin be-coming wetter (Abatzoglou and Kolden 2011). As the climate continues to change, the distribution of vegetation that characterizes the cold and warm desert biomes is predicted to shift northwards in future climates (Rehfeldt et al. 2012). Where the vegetation of the two deserts come into contact, there is a transitional zone comprised of both Mojave and Great Basin species that ranges as wide as 115 km (Beatley 1976; Brussard et al. 1998; Hansen et al. 1999). Understanding distributional change at the Mojave-Great Basin transition zone, will show whether species of the warm deserts, which continue to warm and dry (Dominguez et al. 2010; Abatzoglou and Kolden 2011; Bai et al. 2011), are expanding into the cold desert with less warming (Tang and Arnone 2013), or whether the transition zone behaves as a barrier to northward expansion.

Transition zones between biomes commonly have different avian community dynamics than biome cores (Karanth et al. 2006), but the impact of transition zones on avian distributional changes in response to changing climate has not yet been investigated. At coarse scales, a globally coherent pattern of species shifting their distributions poleward in response to climate change (Parmesan and Yohe 2003; Chen et al. 2011). In the Southwestern deserts, species richness and colonization/extinction rates are higher at the boundaries of each desert than in interior regions (Karanth et al. 2006), which is consistent with macroecological theory (Brown 1984; Shmida and Wilson 1985). A transition zone can persist through a changing climate (Peters et al. 2006) if: (i)

41

biotic factors lag behind climate, such as the leading and trailing-edge range disequilibria in vege-tation response (Svenning and Sandel 2013), and (ii) inherent abiotic factors, such as topography (Gosz 1992) and edaphic factors, contribute to the Mojave-Great Basin transition (Beatley 1975; Hansen et al. 1999). A marked decline was recently identified in the Mojave avifauna (Chapter 2) and barriers to northward expansion could result in range collapse for hot desert species.

For breeding birds of the Mojave and Great Basin Deserts we examined whether the: (i) Mojave and Great Basin avifauna were identifiable as distinct communities historically and whether this distinction remained stable over time; (iilatitude historically and whether species experienced shifts, expansions, or contractions over time; (iii) transition zones between the deserts play a role in limiting community turnover; and (iv) avi-fauna conformed to idealized patterns of community structure and if this nominal classification was retained through time. We predicted that the differing physiogeographic composition of the hot and cold desert biomes and variation in community dynamics found at transition zones will cause hot and cold desert avifauna to be distinguishable from each in other the historic surveys and that idiosyncratic species-level distributional responses will reduce this distinction in the mod-ern surveys. As a corollary, we predicted that historically species would be more strongly delineated into different community types (i.e., have a Clementsian community structure) than now because individualistic distributional responses to climate change would disaggregate this structuring (i.e., a shift towards Gleasonian structure).

Our cross-system comparison enables us to confirm which species responses occur at broader scales, and which are more regionalized and system-specific. We also hypothesized that the declines in Mojave avifauna identified in Chapter 2 would be matched by northward expan-sions of range limits into the Great Basin if the transition zone does not act as a barrier to species-level response. Alternatively, the persistent abiotic factors that comprise the Mojave-Great Basin transition zone and potential lags in vegetation change could act as barriers to avian distributional response to climate change. As a result, we predicted that temporal community turnover would vary as a function of distance from the transition zone, with turnover being more stable over time in the core of each desert than closer to the boundary.

3.2 Methods

3.2.1 Study Area

Historic and modern surveys were conducted at 106 sites for avian diversity. Sites ranged from the

transition with the Columbia Plateau, spanning a latitudinal extent of 33.6° to 41.7° (Figure 3.1). The Mojave and Great Basin are bounded on the east and west by mountain ranges, that sharply circumscribes much of the desert biota (Billings 1990; Brussard et al. 1998). The southern limits of the Mojave gradually transition to the Sonoran Desert (MacMahon 1979), while the Great Basin fades into the Columbia Plateau to the north (Brussard et al. 1998). Temperature and precipitation

42

shifts and corresponding changes in biota clearly defines the boundary between the two deserts. Most sites (102 of 106; 96%) were on federal lands managed by the US National Park Service (NPS), US Bureau of Land Management (BLM), US Forest Service (USFS), and US Fish and Wildlife Ser-vice. Only one site experienced development after historic surveys; resurvey of this site occurred < 1 km away in habitat matching historic descriptions.

3.2.2 Field Surveys

1908 and 1948. Historic observations and associated metadata were transcribed from field notes archived at the Museum of Vertebrate Zoology, at University of California, Berkeley (available online). Photographs, maps, and qualitative descriptions within notebooks allowed identification of historic survey locations with a high degree of accuracy. Previous resurveys using similar meth-ods, estimated spatial error for locating true historic sites averaged 0.7 km (Wieczorek et al. 2004).

precursors to modern line transects (Bibby et al. 2000) and species encounter lists that provide detection/nondetection data. A limited number (n = 22) of historic surveys followed a removal protocol (MacKenzie 2018: 469) and were included in the analysis as such. Sites were selected from historic collecting trips based on the occurrence of dedicated avian sampling and repeated survey effort over multiple days within a single breeding season. A limited number (n = 14) of historic sites were sampled between 11 and 38 times. Because frequently-sampled sites may bias detection probabilities, only the first 10 visits that occurred during a single breeding season were included in the analysis (Socolar et al. 2017).

Modern resurveys occurred between 2013 and 2016 and followed standard avian variable-distance point count methods and protocols established by the Grinnell Resurvey Project (Ralph et al. 1995; Tingley et al. 2012). Resurveys consisted of a transect with 10 point count stations a minimum of 250 m apart. Survey transects were visited three times within a week. Transects routes were chosen to match the topographic extent of the historic surveys and to span the local vegeta-tion heterogeneity described in the historic notes. Abundance data from the modern surveys were reduced to detection/nondetection data for comparison to the historic surveys.

Historic data included 5662 species detections during 445 surveys, with an average of 4.2 surveys per site. Modern data encompassed 5118 species detections during 318 surveys, with an average of 3.0 surveys per site. A total of 162 breeding species were included in the analysis. We excluded nocturnal species and obligate waterbirds because they were not adequately surveyed by our methods. Migrant species observed outside of their known breeding ranges were also excluded due to concerns of closure during the sampling period and our objective of evaluating changes in the breeding community. Species were analyzed based on an updated taxonomic treatment using disjunct geographic ranges or museum specimens that were collected concurrently with the his-

possible by these methods, so they were included in the analysis as their historic species complex: the Sage Sparrow.

43

3.2.3 Modeling Framework

A multispecies occupancy model was employed to explore occupancy and community composi-tion over space and time. A parameterization of a dynamic occupancy model was chosen that ex-plicitly models occupancy probabilities for each time period (Royle and Dorazio 2008: 311; Russell et al. 2009). Species encounters, denoted as ( = 1 for a detection, and = 0 for a nondetection of a species , at site , time period , and survey ), were assumed to be the result of the imperfect observation of the true incidence of a species: ( = 1 if present, and = 0 if absent). The inci-dence of a species at a site was governed by the occupancy probability of a species at that site:

(3.1)

Observations arose from the incidence of a species at a site and the probability detecting that spe-cies, :

(3.2)

where was always zero if the species was absent from the site; otherwise the species was de-tected with the probability . The pattern of detections and nondetections of a species across repeated surveys at a site within a time peribility within the model, assuming closure of that site to immigrations and emigrations.

We built a latitudinal occupancy profile for each species in the historic and modern surveys by fitting covariates to the model of occupancy probability through a logit-link function:

(3.3)

Temporal dependence of occupancy from the historic to modern surveys was included by incor-porating an auto-logistic parameter ( ) in the model of modern occupancy:

(3.4)

Detectability was modeled as a function of the time of year, allowing for detectability of species to vary and peak during the breeding season, and survey era (included as a categorical effect) to ex-plicitly capture any differences in detectability between historic and modern surveys:

(3.5)

The model was fit using JAGS (Plummer 2003) with pre- and post-processing in R v3.4.4 (R Core Team 2017). Species-specific intercept, coefficient, and auto-logistic parameters were drawn from separate community-level hyperdistributions. Means of hyperdistributions were as-signed vague priors and a weakly-informative prior was applied to the variance of the hyperdistri-butions (Gelman et al. 2008; Broms et al. 2016). Forty parallel chains were run. After each chain was fully adapted and an initial burn-in of 10,000 iterations was performed, an additional 32,500 iterations were run with the retention of every 100th sample, resulting in a total of 13,000 samples

44

of the posterior. Convergence was assumed when the Gelman-Rubin statistic was <1.1 for all sto-chastic nodes (Gelman et al. 2014). For JAGS model code, see Appendix B.

3.2.4 Modeling Latitudinal Shifts

We estimated detection-corrected temporal shifts in the latitudinal range limits for each species (Moritz et al. 2008; Tingley et al. 2012; Rowe et al. 2015). Species detections were used to establish potential latitudinal limits in each time period. Across the sites at which an apparent latitudinal shift occurred, we calculated the probability that a species was present at those sites but undetected in the other time period, i.e. the probability of false absence ( ) at those sites:

(3.6)

where is the probability of detecting a species across all surveys performed at a site, and m is the total number of sites within the potential range shift. Shifts were considered statistically significant when the 95% credible intervals (CRI) of

3.2.5 Partitions of Beta Diversity and Transition Zone Dynamics

Temporal change in species assemblages, i.e., beta diversity, was evaluated as a function of distance from the ecoregion boundary between the Great Basin and Mojave Deserts. Change in assemblages are the result of two separate processes: loss of species (i.e., nestedness) and the replacement of species by others (i.e., turnover). Three components of beta diversity were calculated for each pos-terior sample of the true incidence matrix, (i) partitions of beta diversity represented by loss of species, measured by the nested-resultant dissimilarity index (Baselga 2010):

(3.7)

(ii (Simpson 1943; Koleff et al. 2003):

(3.8)

and (iii (Sørensen 1948; Koleff et al. 2003):

(3.9)

45

where a is the number of species present at both time periods, b is the number of species present in historic but not modern surveys, and c is the number of species present in modern but not historic surveys. The importance of proximity to the transition zone between the two deserts to temporal changes in beta diversity was tested using a generalized linear model. Because, we as-sumed the impacts of the transition zone would be strongest near the boundary and fall off towards the core of the biome, the effect of distance was fit as negative exponential distance decay.

3.2.6 Elements of Metacommunity Structure

We used detection-corrected Elements of Metacommunity Structure (DECEMS; Mihaljevic et al. 2015) to categorize the structure of the avian community in both time periods across the entire study area. Error from imperfect detection was accounted for in the categorization of metacommu-nity structure by replacing the uncorrected site-by- -employed in the Elements of Metacommunity Structure (EMS) framework developed by Leibold and Mikkelson (2002) ) estimated by the multispecies occu-pancy model. Prior to analysis, the incidence matrix was ordinated along the species axis using reciprocal averaging and the site axis was sorted according to latitude, following recommendations by Dallas et al. (2016). Community structure was then classified using three stepwise tests for pat-terns in species co-occurrences: coherence, species replacements, and boundary clumping. These three tests were used to identify six classical structures recognized in community ecology (Leibold and Mikkelson 2002): random, checkerboard, nested, evenly spaced, Gleasonian, and Clementsian.

-nificantly coherent (i.e., nonrandom/noncepplacements (Presley et al. 2010). Coherence was tested based on counting embedded absences in the ordinated matrix compared to the distribution of embedded absences generated from 1000 simulated matrices. Replacement, the number of times that one species is replaced by another across sites, was also tested against the distribution generated by simulated matrices. Boundary clumping and its oppositeindex. Details of the underlying community ecological theory and methods are covered by Leibold and Mikkelson (2002) and Presley et al. (2010). DECEMS metrics were calculated in R using meth-ods from Dallas et al. (2016) the packages metacom (Dallas 2014) and vegan (Oksanen et al. 2018). To reduce computational time, t -incidence matrix, , i.e., a random draw of 1,000 samples from the full posterior (Mihaljevic et al. 2015).

46

3.3 Results

3.3.1 Avifauna and Species-level Changes

Across both avifaunas, modern occupancy ( = 0.03, 95% CRI: 0.02 - 0.04) was significantly lower than historic occupancy ( = 0.15, 95% CRI: 0.12 - 0.19). Species present at a site during modern surveys were very likely to be encountered at least once ( = 0.96; 95% CRI: 0.95 - 0.97). Site-level detection was significantly lower for historic surveys but was still relatively high ( = 0.75; 95% CRI: 0.70 - 0.80).

No species significantly increased in one desert and declined in the other, or vice versa (Figure 3.2), although there was variation in the amount of occupancy change a species experi-enced. Overall, occupancy did not change significantly in either desert for the majority of species (54.3%). A substantial proportion of species (39.5%) declined in one (Mojave: n = 9; Great Basin: n = 31) or both (n The Great Basin had 8 other species increasing significantly: American Crow, Broad-tailed Hum-mingbird, Cassin's Finch, Chukar, Gray Vireo, Northern Rough-winged Swallow, Red-naped Sap-sucker, and Spotted Towhee, while the Mojave had only one other: the Common Raven.

Figure 3.2 Occupancy Change by Desert Occupancy change by species (n = 162) in the Mojave and Great Basin Deserts. Species whose occupancy changed the same in both regions are symbolized with circles: significant increases (blue), significant declines (red), no significant change (gray). Species that dis-played significant changes in the Mojave but not the Great Basin are symbolized with squares and vice versa with triangles: significant increases (teal) and significant declines (orange).

47

3.3.2 Shifting Ranges

More species had occupancies structured by latitude in the historic (40.7%) than modern surveys (22.8%). In both time periods nearly equal proportion of species had occupancies which increased with latitude as decreased with latitude (historic: 19.1% and 21.6%, modern: 10.5% and 12.3%, respectively; Figure 3.3). The antithetical response of subsets of the community meant latitude was not a significant predictor of occupancy across the community as a whole in the historic ( = -0.03, 95% CRI: -0.27 - 0.20) or modern time period ( = 0.02, 95% CRI: -0.14 - 0.19). The structuring identified by the occupancy model generally reaffirms the traditional classification of the Mojave and Great Basin avifauna as identified by Miller (1951), Behle (1978), and Udvardy (1963). Of members of the traditional avifauna with significant latitudinal structuring, all Mojave species had occupancies that decreased moving northwards (i.e., the red plotted lines in Figure 3.3), and most species of the Great Basin (73.6%) had occupancies that increased from the southern parts of their range (i.e., the blue plotted lines in Figure 3.3).

Figure 3.3 Species' Occupancy Probability by Latitude The relationship of species occupancy probability with latitude in (top panel) the historic (n = 66) and (bottom panel) the modern time periods (n = 37). Only species with signifi-cant responses (i.e., 95% credible interval did not overlap with zero) are displayed for vis-ual clarity. Species that significantly declined with latitude are in red, species that in-creased are in blue. The dotted line is the average latitude of the boundary between the Mojave and Great Basin Deserts (37.2°).

48

Directionality of changes in latitudinal limits were idiosyncratic across the community. We were able to evaluate range shifts of 136 species. Across these species, we detected 100 significant latitudinal shifts, which represented change in 36.8% of the species-specific historic range limits (Figure 3.4). We found no significant range change for 56 species (41.2%). Of the remaining 80 species, an equal number contracted their northern or southern limits (n = 37). However, species were more likely expand their northern limits (n = 20) than their southern limits (n = 6).

If species are subdivided into groups based on their historical latitudinal extent, a more nuanced picture of range shifts emerges (Figure 3.5). Species were split into four groups: those whose historic ranges were predominantly in the Mojave Desert (southern: 75% of distribution < 37.2° latitude), predominately in the Great Basin (northern: 75% of distribution > 37.2° latitude),

Figure 3.4 Latitudinal Range Limit Shifts. Latitudinal ranges of Mojave and Great Basin bird species (n = 136) arranged by group and mean latitude of the his-toric range. Black bars show historic range, with significant expansions from the historic time period in green (non-significant expansions in white) and significant contractions in red (non-significant shifts in gray). Species were splitinto groups based on the extent of their historical ranges: southern (75% of distribution < 37.2° latitude), northern (75% of distribution > 37.2° latitude), wide-ranging (distributional extent covered >75% of study area), and mid-lati-tude (i.e., any species not meeting the aforementioned criteria). The dotted line is the average latitude of the boundary between the Mojave and Great Basin Deserts (37.2°).

49

wide-ranging (distributional extent covered >75% of study area), and spanning the mid-latitudes (i.e., any species not meeting the aforementioned criteria). Save for one significant northern range expansion, wide-ranging species only contracted their ranges or remained stable, however, the lack of observable expansions is an obvious corollary of their historic range limits exceeding the bounds of the study area (Figure 3.4). Just over half (51.6%) of wide-ranging species con-tracted one or both of their range limits. Expansion of the southern limits of Mojave species were similarly obscured by artificial truncation by the study area, and only contractions of southern limits (23.3%) or stability were observed for this group. There were an equal number of range contractions and expansions for species in the mid-latitudes, and, notably, only one Great Basin species expanded its limits southward. The proportion of expanding and contracting northern limits were relatively more even across all three groups: Mojave species were 1.6 times more likely to expand their upper limits, while mid-latitude species were more likely to contract theirs (1.15 and 1.22 times more likely, respectively).

3.3.3 Community Turnover and the Transition Zone

The composition of species at the survey sites became more disparate from each other over time. Although, the average dissimilarity across all sites did not differ significantly ( = 0.61 95% CRI: 0.54-0.69; = 0.70 95% CRI: 0.63-0.80), a majority of sites (n = 74; 69.8%) became significantly more distinct in their composition. There was no discernible spatial pattern in how communities change at the site level over time. Sites that were closer to the transition from the Mojave to the Great Basin Desert showed no distinguishable differences in community simi-larity between the historic and modern surveys than sites that were more distant (Figure 3.6). This

Figure 3.5 Direction of Latitudinal Shifts Proportion of latitudinal range limit shifts by species whose historic ranges were predom-inately in the Mojave (southern), the Great Basin (northern), and spanning the mid-lati-tudes: significant expansions (green), significant contractions (red), or no significantchange (gray).

50

relationship held across beta diversity measures as a whole ( : = 0.07) and its two par-titions: species loss ( : = 0.07) and species replacements ( : = 0.07).

3.3.4 A Shift in Community Structure

Mirroring the loss of latitudinally-structured occupancies and the idiosyncratic range shifts of in-dividual species, the avifaunas of the two deserts were less strongly structured along a latitudinal gradient presently than they were in the past. Historically, the metacommunity had a Clement-sian/quasi-Clementsian structure (52.7% and 40.9% of respectively). This shifted to a quasi-Clementsian majority (73.6%), quasi-nested minority (20.2%) modern metacommunity (Table 3.1). Except for 0.04% of in the historic period that was classified as random, the meta-community remained unequivocally coherent with significant boundary clumping. The increasing dominance of quasi-structures, indicates a difference between the past and present in the pattern of compositional change along the latitudinal gradient, with species losses and replacements be-coming more random and less strongly structured from north to south (Presley et al. 2010).

Figure 3.6 Change in Community Composition and the Ecoregion Transition Site-level temporal dissimilarity in community composition compared to site-level distance from the Mojave-Great Basin Desert boundary. The importance of proximity to the transition zone between the two deserts was tested using a nega-tive exponential distance decay model fit via a generalized linear model. Shadedregion represent 95% confidence intervals around the estimated mean.

51

3.4 Discussion The results from a cross-biome analysis provide strong evidence for a redistribution of breeding bird species across the Mojave and Great Basin Deserts. Our results show that the transition zone that has historically separated the Mojave and Great Basin avifauna (Figure 3.3; Miller 1951; Udvardy 1963; Behle 1978) no longer strongly defines the community. Significant and biologically relevant range shifts have occurred in both northern and southern species limits, and species oc-cupancy and the avifauna, as a whole, have lost latitudinal structuring. However, this community reshuffling lacks a consistent pattern, and as of now, only two species traditionally associated with the Mojave, the Gray Vireo and Northern Rough-winged Swallow, have significantly increased their occupancy in the Great Basin (Figure 3.2).

3.4.1 A Redistribution of Species

Although species were more likely to expand their northern limits than southern (Figure 3.4 and Figure 3.5), there was not an overwhelming signal of poleward movement in the birds of the de-serts. Substantial heterogeneity in distributional shifts were prevalent across both wide-ranging and latitudinally-restricted species and significant and biologically relevant shifts happened in both northern and southern limits. Poleward shifts have been documented for many taxa (Parmesan and Yohe 2003; Hickling et al. 2006; Chen et al. 2011), and latitudinal shifts have been particularly well-described in bird communities globally (Hitch and Leberg 2007; Zuckerberg et al. 2009; Moreno-Rueda 2010; Brommer et al. 2012; Virkkala and Lehikoinen 2014; Gillings et al. 2015; Lehikoinen and Virkkala 2015; Massimino et al. 2015; Tayleur et al. 2015; Wu and Zhang 2015; Hovick et al. 2016; Wu and Shi 2016; Huang et al. 2017). The dominant direction reported for most studies of avian distributional shifts has been in the poleward direction, though this pattern does appear to be system- and scale-dependent in some cases (see Currie and Venne, 2017; Vanderwal et al., 2013). Correcting for detection when investigating changes in range limits is particularly important because nondetections can underestimate range boundaries and artificially magnify range dynamics (Doherty et al. 2003). A few of the earlier studies identifying poleward movement failed to account for changes in detectability in survey methods over time (Thomas and Lennon

Table 3.1. Metacommunity Structure and Associated Statistics. The percentage of the pseudo-posterior draws, , classified as each community type is presented along with the median z-scores of coherence and turnover measures. Structure Type z-scores

Clementsian Quasi-

Clementsian Quasi-Nested Random

Coherence

Species replacements

Historic 52.70% 40.90% 6.00% 0.40% 6.42 -2.06 Modern 6.20% 73.60% 20.20% 5.71 -0.58

52

1999; Brommer 2004; Zuckerberg et al. 2009), and when reanalyzed using detection-corrected methods, the unidirectional signal was either lost entirely or greatly diminished (Kujala et al. 2012).

Unidirectional movement along a north-south gradient is a simplified view of spatial shift-ing, and studies that have sought to identify multi-directional change have found movement to-wards all points of the compass (Vanderwal et al. 2013; Gillings et al. 2015; Lehikoinen and Virk-kala 2015; Gaüzère et al. 2016). However, in the Mojave and Great Basin it is a suitable proxy for movement across the east-west boundary between the biomes (Figure 3.1). Species on both sides of boundary and those spanning the divide expanded and contracted their northern limits in fairly equal proportions, as did mid-latitude species with their southern limits (Figure 3.4). However, Great Basin species, except for one, did not expand their ranges southward which may suggest there is some limit to the movement of species through time. The idiosyncratic nature of latitudinal shifts for desert birds is also consistent with the substantial heterogeneity found among altitudinal shifts a complementary mechanism to poleward shifts observed across many montane taxa (Rapacciuolo et al. 2014): birds (Tingley et al. 2012), plants (Crimmins et al. 2011), mammals (Rowe et al. 2015), and butterflies (Forister et al. 2010).

How occupancy changed for individual species was consistent across the deserts: if a spe-cies was declining in one desert it was either in decline or had no discernible change in the other (Figure 3.2). Relatively few species had significant increases in occupancy (6.2%), and a substantial portion were in decline in either desert (39.5%). Integrating range shifts with occupancy change provides richer information on the status of individual species and can elucidate whether shifts might compensate for population declines. Generally, occupancy dynamics paralleled range dy-namics. Of the species in significant decline in one or both deserts that were included in the range analysis (n = 55), the majority experienced range contractions (63.6%) or had stable limits (30.9%), with only a few expanding their ranges (5.5%). Of the species that increased in occupancy (n = 7), the majority expanded their limits (71.4%), however, some experienced contractions (28.6%). Spe-cies without a trend in occupancy (n = 74) also tended to have stable (51.4%) or expanding (20.3%) ranges, with a smaller proportion of species contracting their ranges (28.4%). Range contractions concomitant with population declines is a common pattern among bird species (Mehlman 1997; Rodríguez 2002; Ralston et al. 2017), which means either may serve as suitable indicators of species of conservation concern.

3.4.2 Communities in Flux

The desert avifauna is more weakly structured presently than it was in the past. Historically, bird co-occurrences formed multiple distinct communities that replaced each other from north to south. The transition from community to community has become more indistinct overtime (Table 3.1). The highly idiosyncratic nature of range shifts across species (Figure 3.4) did not dismantle co-occurrences sufficiently to transform the metacommunity into one whose arrangement was driven by individualistic response (i.e., a Gleasonian community). However, there are limitations

53

in the EMS framework to quantifying more nuanced shifts in a community from Clementsian to-wards a Gleasonian one. Because boundary clumping which underlies the quantitative distinc-tion of these community types (Morisita 1962), we are unable to investigate whether there was a shift in the degree or magnitude of clumping over time, only that species arrangement along that gradient did not become entirely random (Keith et al. 2011). Only a few studies have investigated change in metacommunity structure over longer time periods (>20 years; Keith et al. 2011; Newton et al. 2011; Diaz et al. 2013; Bonthoux and Balent 2015; Alignier 2018), and every community studied retained a Clementsian (or quasi-Clement-sian) structure through time. Moreover, this nominal structure was invariant to change in beta, gamma, and alpha diversity (Keith et al. 2011; Bonthoux and Balent 2015; Alignier 2018), which indicates that community structure may be a poor indicator of conservation status of a region.

We hypothesized that if the transition between the Mojave and Great Basin Deserts could behave as a persistent barrier for avifauna and that it would coincide with higher levels of commu-nity turnover near the boundaries than in the core habitat of each desert. Macroecological theory posits that: (i) abundance and occupancy decline towards range peripheries (Brown 1984; Maurer and Brown 1989; Gaston 1990; but see Sagarin et al. 2006 for a critique of the abundant-center hypothesis); (ii) higher rates of colonizations and extinctions coincide with lower abundance and occupancy; and (iii) ecotones and habitat boundaries are often conterminous with species range limits (Delcourt and Delcourt 1992; Kark and Rensburg 2006). Increased extinctions and coloni-zations near range margins and at transitions between ecological regions have been identified in many bird species (Mehlman 1997; Doherty et al. 2003; Karanth et al. 2006; La Sorte and Thomp-son 2007; Royle and Kéry 2007) sition zone was discernible (Figure 3.6). The consistency in turnover in space may further indicate that the boundary did not behave as a substantial barrier structuring the range limits of species over time, which is also consistent with the loss of latitudinal structuring of the community (Table 3.1) and of species occupancies (Figure 3.3).

3.4.3 Conservation Implications

Change in the avian communities of the Mojave and Great Basin Deserts over the past century is defined by species redistributions and degradation of historic community structure. The signal of decline identified in the Mojave (Chapter 2) does not appear to be offset by compensatory range expansions into the Great Basin. Instead, most species with declining occupancies had correspond-ing range contractions, which highlights that species in this category are of special conservation concern. Range stability in a changing environment, however, does not necessarily signify that a species is in trouble. Some species appear to be currently coping in situ, potentially through mech-anisms such as by advancing their breeding pheonologies (Socolar et al. 2017). The population declines and range contractions found in both range-restricted and wide-ranging species necessi-tates that future conservation efforts be holistic in their approach and have a broader focus beyond rare and endangered species. Redistributions are bringing new species into contact and driving the

54

loss of historic communities and genesis of new ones. The ecological consequences of this will require additional monitoring and flexible management into the future (Bonebrake et al. 2018).

55

Abatzoglou JT, Kolden CA. 2011. Climate change in western US deserts: Potential for increased

wildfire and invasive annual grasses. Rangeland Ecology & Management. 64(5):471 478. doi:10.2111/REM-D-09-00151.1.

Ahumada JA, Silva CEF, Gajapersad K, Hallam C, Hurtado J, Martin E, McWilliam A, Mugerwa

mammals: Data from a global camera trap network. Philosophical Transactions of the Royal Society B: Biological Sciences. 366(1578):2703 2711. doi:10.1098/rstb.2011.0115.

Aing C, Halls S, Oken K, Dobrow R, Fieberg J. 2011. A Bayesian hierarchical occupancy model for track surveys conducted in a series of linear, spatially correlated, sites. Journal of Applied Ecology. 48(6):1508 1517. doi:10.1111/j.1365-2664.2011.02037.x.

BO. 2017. Mapping evaporative water loss in desert passerines reveals an expanding threat of lethal dehydration. Proceedings of the National Academy of Sciences. 114(9):2283 2288. doi:10.1073/pnas.1613625114.

Alignier A. 2018. Two decades of change in a field margin vegetation metacommunity as a result of field margin structure and management practice changes. Agriculture, Ecosystems & Environment. 251:1 10. doi:10.1016/j.agee.2017.09.013.

Alldredge MW, Simons TR, Pollock KH. 2007. Factors affecting aural detections of songbirds. Ecological Applications. 17(3):948 955. doi:10.1890/06-0685.

predict recent shifts at expanding range edges? Ecology Letters. 14(7):677 689. doi:10.1111/j.1461-0248.2011.01620.x.

Bai Y, Scott TA, Chen W, Minnich RA, Chang AC. 2011. Long-term variation in soil temperature of the Mojave Desert, southwestern USA. Climate Research. 46(1):43 50. doi:10.3354/cr00970.

Bailey LL, Reid JA, Forsman ED, Nichols JD. 2009. Modeling co-occurrence of northern spotted and barred owls: Accounting for detection probability differences. Biological Conservation. 142(12):2983 2989. doi:10.1016/j.biocon.2009.07.028.

Barnes MD, Craigie ID, Harrison LB, Geldmann J, Collen B, Whitmee S, Balmford A, Burgess ND, Brooks T, Hockings M, et al. 2016. Wildlife population trends in protected areas

56

predicted by national socio-economic metrics and body size. Nature Communications. 7:12747. doi:10.1038/ncomms12747.

Baselga A. 2010. Partitioning the turnover and nestedness components of beta diversity. Global Ecology and Biogeography. 19(1):134 143. doi:10.1111/j.1466-8238.2009.00490.x.

Beatley JC. 1975. Climates and vegetation pattern across the Mojave/Great Basin Desert transition of Southern Nevada. American Midland Naturalist. 93(1):53 70. doi:10.2307/2424105.

Beatley JC. 1976. Rainfall and fluctuating plant populations in relation to distributions and numbers of desert rodents in Southern Nevada. Oecologia. 24(1):21 42. doi:10.2307/4215266.

Beck J, Holloway JD, Schwanghart W. 2013. Undersampling and the measurement of beta diversity. Methods in Ecology and Evolution. 4(4):370 382. doi:10.1111/2041-210x.12023.

Behle WH. 1978. Avian biogeography of the Great Basin and Intermountain region. Great Basin Naturalist Memoirs. 2:55 80. https://www.jstor.org/stable/23376559.

Berggren Å, Björkman C, Bylund H, Ayres MP. 2009. The distribution and abundance of animal populations in a climate of uncertainty. Oikos. 118(8):1121 1126. doi:10.1111/j.1600-0706.2009.17558.x.

Bergstrom DM, Bricher PK, Raymond B, Terauds A, Doley D, McGeoch MA, Whinam J, Glen M, Yuan Z, Kiefer K, et al. 2015. Rapid collapse of a sub-Antarctic alpine ecosystem: The role of climate and pathogens. Journal of Applied Ecology. 52(3):774 783. doi:10.1111/1365-2664.12436.

Bibby CJ, Burgess ND, Hill DA, Mustoe S. 2000. Bird Census Techniques. 2nd ed. San Diego, CA: Elsevier.

Billings WD. 1990. The mountain forests of North America and their environments. In: Osmond CB, Hidy GM, Pitelka LF, editors. Plant Biology of the Basin and Range. Berlin, DE: Springer. p. 47 86.

Blomberg SP, Garland T, Ives AR, Crespi B. 2003. Testing for phylogenetic signal in comparative data: Behavioral traits are more labile. Evolution. 57(4):717 745. doi:10.1554/0014-3820(2003)057[0717:TFPSIC]2.0.CO;2.

57

Boarman WI, Camp RJ, Hagan M, Deal W. 1995. Raven abundance at anthropogenic resources in the western Mojave Desert, California. Riverside, CA: National Biological Service Report to Edwards Air Force Base. [accessed 2018 Jan 1]. http://pubs.er.usgs.gov/publication/96735.

Bonebrake TC, Brown CJ, Bell JD, Blanchard JL, Chauvenet A, Champion C, Chen IC, Clark TD, Colwell RK, Danielsen F, et al. 2018. Managing consequences of climate-driven species redistribution requires integration of ecology, conservation and social science. Biological Reviews. 93:284 305. doi:10.1111/brv.12344.

Bonthoux S, Balent G. 2015. Bird metacommunity processes remain constant after 25 years of landscape changes. Ecological Complexity. 21:39 43. doi:10.1016/j.ecocom.2014.12.002.

Bried JT, Hager BJ, Hunt PD, Fox JN, Jensen HJ, Vowels KM. 2012. Bias of reduced-effort community surveys for adult Odonata of lentic waters. Insect Conservation and Diversity. 5(3):213 222. doi:10.1111/j.1752-4598.2011.00156.x.

Brommer JE. 2004. The range margins of northern birds shift polewards. Annales Zoologici Fennici. 41(2):391 397. doi:10.1111/bij.12574.

Brommer JE, Lehikoinen A, Valkama J. 2012. The breeding ranges of central European and arctic bird species move poleward. PLoS ONE. 7(9):e43648. doi:10.1371/journal.pone.0043648.

Broms KM, Hooten MB, Fitzpatrick RM. 2016. Model selection and assessment for multi-species occupancy models. Ecology. 97(7):1759 1770. doi:10.1890/15-1471.1.

ace JB, Keeley JE, DiTomaso JM, Hobbs RJ, Pellant M, Pyke D. 2004. Effects of invasive alien plants on fire regimes. BioScience. 54(7):677 688.

Brooks ML, Pyke DA. 2001. Invasive plants and fire in the deserts of North America. In: Galley KEM, Wilson TP, editors. Proceedings of the Invasive Species Workshop: The Role of Fire in the Control and Spread of Invasive Species. Vol. 11. Tallahassee, FL: the First National Congress on Fire Ecology, Prevention, and Management. p. 1 14.

Brown JH. 1984. On the relationship between abundance and distribution of species. The American Naturalist. 124(2):255 279. doi:10.1086/284267.

Brown JH. 1995. Macroecology. Chicago, IL: University of Chicago Press.

Brussard PF, Charlet DA, Dobkin DS. 1998. Great Basin-Mojave Desert region. In: Mac MJ,

Resources. Vol. 2. Reston, VA: USGS, USDOI. p. 505 542.

58

Buckland ST, Anderson DR, Burnham KP, Laake JL, Thomas L. 2001. Introduction to Distance Sampling: Estimating Abundance of Biological Populations. Oxford, UK: Oxford University Press.

Buckland ST, Borchers DL, Johnston A, Henrys PA, Marques TA. 2007. Line transect methods for plant surveys. Biometrics. 63(4):989 998. doi:10.1111/j.1541-0420.2007.00798.x.

Burnham KP, Anderson DR. 2002. Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach. 2nd ed. New York, USA: Springer.

Burnham KP, Overton WS. 1979. Robust estimation of population size when capture probabilities vary among animals. Ecology. 60(5):927 936. doi:10.2307/1936861.

Burton AC, Sam MK, Balangtaa C, Brashares JS. 2012. Hierarchical multi-species modeling of carnivore responses to hunting, habitat and prey in a West African protected area. PLoS ONE. 7(5):e38007. doi:10.1371/journal.pone.0038007.

Cameron DR, Cohen BS, Morrison SA. 2012. An approach to enhance the conservation-compatibility of solar energy development. PLoS ONE. 7(6):e38437. doi:10.1371/journal.pone.0038437.

Cardinale BJ, Duffy JE, Gonzalez A, Hooper DU, Perrings C, Venail P, Narwani A, Mace GM, Tilman D, Wardle DA, et al. 2012. Biodiversity loss and its impact on humanity. Nature. 486(7401):59 67. doi:10.1038/nature11148.

Celeux G, Forbes F, Robert CP, Titterington DM. 2006. Deviance information criteria for missing data models. Bayesian Analysis. 1(4):651 673. doi:10.1214/06-BA122.

Chandler RB, King DI, Raudales R, Trubey R, Chandler C, Arce Chávez VJ. 2013. Small-scale land-sparing approach to conserving biological diversity in tropical agricultural landscapes. Conservation Biology. 27(4):785 795. doi:10.1111/cobi.12046.

Chao A. 1984. Nonparametric estimation of the number of classes in a population. Scandinavian Journal of Statistics. 11(4):265 270. doi:10.2307/4615964.

Chao A. 1987. Estimating the population size for capture-recapture data with unequal catchability. Biometrics. 43(4):783 791. doi:10.2307/2531532.

Chao A, Chazdon RL, Colwell RK, Shen TJ. 2005. A new statistical approach for assessing similarity of species composition with incidence and abundance data. Ecology Letters. 8(2):148 159. doi:10.1111/j.1461-0248.2004.00707.x.

59

Chao A, Chazdon RL, Colwell RK, Shen TJ. 2006. Abundance-based similarity indices and their estimation when there are unseen species in samples. Biometrics. 62(2):361 371. doi:10.2307/3695855.

Chao A, Gotelli N, Hsieh TC, Sander E, Ma KH, Colwell RK, Ellison AM. 2013. Rarefaction and extrapolation with Hill numbers: A framework for sampling and estimation in species diversity studies. Ecological Monographs. doi:10.1890/13-0133.1.

unseen species in sample. Environmental and Ecological Statistics. 10(4):429 443. doi:10.1023/A:1026096204727.

Cheal AJ, Emslie M, MacNeil MA, Miller I, Sweatman H. 2012. Spatial variation in the functional characteristics of herbivorous fish communities and the resilience of coral reefs. Ecological Applications. 23(1):174 188. doi:10.1890/11-2253.1.

Chen G, Kéry M, Plattner M, Ma K, Gardner B. 2013. Imperfect detection is the rule rather than the exception in plant distribution studies. Journal of Ecology. 101(1):183 191. doi:10.1111/1365-2745.12021.

Chen IC, Hill JK, Ohlemüller R, Roy DB, Thomas CD. 2011. Rapid range shifts of species associated with high levels of climate warming. Science. 333(6045):1024 1026. doi:10.1126/science.1206432.

Chesson P, Gebauer RLE, Schwinning S, Huntly N, Wiegand K, Ernest MSK, Sher A, Novoplansky A, Weltzin JF. 2004. Resource pulses, species interactions, and diversity maintenance in arid and semi-arid environments. Oecologia. 141(2):236 253. doi:10.1007/s00442-004-1551-1.

Clark JS, Ferraz G, Oguge N, Hays H, DiCostanzo J. 2005. Hierarchical Bayes for structured, variable populations: from recapture data to life-history prediction. Ecology. 86(8):22322244. doi:10.1890/04-1348.

Crawford CS. 1981. Biology of Desert Invertebrates. New York, NY: Springer-Verlag.

Crimmins SM, Dobrowski SZ, Greenberg JA, Abatzoglou JT, Mynsberge AR. 2011. Changes in climatic water balance drive 331(6015):324 327. doi:10.1126/science.1199040.

Cruz-McDonnell KK, Wolf BO. 2016. Rapid warming and drought negatively impact population size and reproductive dynamics of an avian predator in the arid Southwest. Global Change Biology. 22(1):237 253. doi:10.1111/gcb.13092.

60

Cunningham SJ, Martin RO, Hockey PA. 2015. Can behaviour buffer the impacts of climate change on an arid-zone bird? Ostrich. 86(1 2):119 126. doi:10.2989/00306525.2015.1016469.

Currie DJ, Venne S. 2017. Climate change is not a major driver of shifts in the geographical distributions of North American birds. Global Ecology and Biogeography. 26(3):333 346. doi:10.1111/geb.12538.

Cutler TL, Morrison ML. 1998. Habitat use by small vertebrates at two water developments in southwestern Arizona. The Southwestern Naturalist. 43(2):155 162.

Dallas T. 2014. metacom: An R package for the analysis of metacommunity structure. Ecography. 37(4):402 405. doi:10.1111/j.1600-0587.2013.00695.x.

Dallas TA, Kramer AM, Zokan M, Drake JM. 2016. Ordination obscures the influence of environment on plankton metacommunity structure. Limnology and Oceanography Letters. 1(1):54 61. doi:10.1002/lol2.10028.

De Graaf RM, Tilghman NG, Anderson SH. 1985. Foraging guilds of North American birds. Environmental Management. 9(6):493 536. doi:10.1007/BF01867324.

Delcourt PA, Delcourt HR. 1992. Ecotone dynamics in space and time. In: Landscape Boundaries. New York, NY: Springer. (Ecological Studies). p. 19 54.

DeWan A, Zipkin E. 2010. An integrated sampling and analysis approach for improved biodiversity monitoring. Environmental Management. 45(5):1223 1230. doi:10.1007/s00267-010-9457-7.

Diaz A, Keith SA, Bullock JM, Hooftman DAP, Newton AC. 2013. Conservation implications of long-term changes detected in a lowland heath plant metacommunity. Biological Conservation. 167:325 333. doi:10.1016/j.biocon.2013.08.018.

Dodd CK, Dorazio RM. 2004. Using counts to simultaneously estimate abundance and detection probabilities in a salamander community. Herpetologica. 60(4):468 478. doi:10.1655/03-60.

Doherty PF, Boulinier T, Nichols JD. 2003. Local extinction and turnover rates at the edge and ici Fennici. 40(2):145 153.

https://www.jstor.org/stable/23736521.

Dominguez F, Cañon J, Valdes J. 2010. IPCC-AR4 climate simulations for the Southwestern US: The importance of future ENSO projections. Climatic Change. 99(3):499 514. doi:10.1007/s10584-009-9672-5.

61

Dorazio RM. 2007. On the choice of statistical models for estimating occurrence and extinction from animal surveys. Ecology. 88(11):2773 2782. doi:10.1890/07-0006.1.

Dorazio RM, Gotelli NJ, Ellison AM. 2011. Modern methods of estimating biodiversity from presence-absence surveys. In: Grillo O, Venora G, editors. Biodiversity Loss in a Changing Planet. InTech. p. 277 302.

Dorazio RM, Kéry M, Royle JA, Plattner M. 2010. Models for inference in dynamic metacommunity systems. Ecology. 91(8):2466 2475. doi:10.1890/09-1033.1.

Dorazio RM, Martin J, Edwards HH. 2013. Estimating abundance while accounting for rarity, correlated behavior, and other sources of variation in counts. Ecology. 94(7):1472 1478. doi:10.1890/12-1365.1.

Dorazio RM, Royle JA. 2005. Estimating size and composition of biological communities by modeling the occurrence of species. Journal of the American Statistical Association. 100(470):389 398. doi:10.1198/016214505000000015.

Dorazio RM, Royle JA, Söderström B, Glimskär A. 2006. Estimating species richness and accumulation by modeling species occurrence and detectability. Ecology. 87(4):842 854. doi:10.1890/0012-9658(2006)87[842:ESRAAB]2.0.CO;2.

Easterday K, McIntyre P, Kelly M. 2018. Land ownership and 20th century changes to forest structure in California. Forest Ecology and Management. 422:137 146. doi:10.1016/j.foreco.2018.04.012.

Ellison AM. 2010. Partitioning diversity. Ecology. 91(7):1962 1963. doi:10.1890/09-1692.1.

Esque TC, Webb RH, Wallace CSA, van Riper C, McCreedy C, Smythe L. 2013. Desert fires fueled by native annual forbs: Effects of fire on communities of plants and birds in the lower Sonoran Desert of Arizona. The Southwestern Naturalist. 58(2):223 233. doi:10.1894/0038-4909-58.2.223.

Ferrier S, Guisan A. 2006. Spatial modelling of biodiversity at the community level. Journal of Applied Ecology. 43(3):393 404. doi:10.2307/3838442.

Fitzpatrick MC, Preisser EL, Ellison AM, Elkinton JS. 2009. Observer bias and the detection of low-density populations. Ecological Applications. 19(7):1673 1679. doi:10.1890/09-0265.1.

Fleishman E, Murphy DD, Sada DW. 2006. Effects of environmental heterogeneity and disturbance on the native and non-native flora of desert springs. Biological Invasions. 8(5):1091 1101. doi:10.1007/s10530-005-7564-9.

62

Flint LE, Flint AL, Thorne JH, Boynton R. 2013. Fine-scale hydrologic modeling for regional landscape applications: The California Basin Characterization Model development and performance. Ecological Processes. 2(1):1 21. doi:10.1186/2192-1709-2-25.

Forister ML, McCall AC, Sanders NJ, Fordyce JA, ThornAM. 2010. Compounded effects of climate change and habitat alteration shift patterns of butterfly diversity. Proceedings of the National Academy of Sciences. 107(5):2088 2092.

hall CD, Zook J, Kremen C, Hadly EA, Daily GC. 2014. Loss of avian phylogenetic diversity in Neotropical agricultural systems. Science. 345(6202):1343 1346. doi:10.1126/science.1254610.

Furnas BJ, McGrann MC. 2018. Using occupancy modeling to monitor dates of peak vocal activity for passerines in California. The Condor. 120(1):188 200. doi:10.1650/CONDOR-17-165.1.

Gaston KJ. 1990. Patterns in the geographical ranges of species. Biological Reviews. 65(2):105129. doi:10.1111/j.1469-185X.1990.tb01185.x.

Gaüzère P, Princé K, Devictor V. 2016. Where do they go? The effects of topography and habitat diversity on reducing climatic debt in birds. Global Change Biology. 23(6):2218 2229. doi:10.1111/gcb.13500.

Gelfand AE, Schmidt AM, Wu S, Silander JA, Latimer A, Rebelo AG. 2005. Modelling species diversity through species level hierarchical modelling. Journal of the Royal Statistical Society: Series C (Applied Statistics). 54(1):1 20. doi:10.1111/j.1467-9876.2005.00466.x.

Gelman A, Carlin JB, Stern HS, Rubin DB. 2014. Bayesian Data Analysis. 3rd ed. Boca Raton, FL: Chapman & Hall/CRC.

Gelman A, Jakulin A, Pittau MG, Su YS. 2008. A weakly informative default prior distribution for logistic and other regression models. The Annals of Applied Statistics. 2(4):1360 1383. doi:10.1214/08-aoas191.

Gelman A, Meng XL, Stern H. 1996. Posterior predictive assessment of model fitness via realized discrepancies. Statistica Sinica. 6(4):733 760. doi:10.2307/24306036.

Gillings S, Balmer DE, Fuller RJ. 2015. Directionality of recent bird distribution shifts and climate change in Great Britain. Global Change Biology. 21(6):2155 2168. doi:10.1111/gcb.12823.

63

Gosz JR. 1992. Ecological functions in a biome transition zone: Translating local responses to broad-scale dynamics. In: Hansen AJ, di Castri F, editors. Landscape Boundaries: Consequences for Biotic Diversity and Ecological Flows. New York, NY: Springer. p. 55 75.

Grant PR, Grant BR. 1987. The extraordinary El Niño event of 1982-finches on Isla Genovesa, Galápagos. Oikos. 49(1):55 66. doi:10.2307/3565554.

Griffis-Kyle KL. 2016. Physiology and ecology to inform climate adaptation strategies for desert amphibians. Herpetological Conservation and Biology. 11(3):563 582.

Hansen DJ, Ostler WK, Hall DB. 1999. The transition from Mojave Desert to Great Basin Desert on the Nevada Test Site. In: Mcarthur ED, Ostler WK, editors. Proceedings: Shrubland Ecotones. Fort Collins, CO: USDA FS, Rocky Mountain Research Station. (RMRS-P-11). p. 148 158.

Hargrove L, Rotenberry JT. 2011a. Breeding success at the range margin of a desert species: Implications for a climate-induced elevational shift. Oikos. 120(10):1568 1576. doi:10.1111/j.1600-0706.2011.19284.x.

Hargrove L, Rotenberry JT. 2011b. Spatial structure and dynamics of breeding bird populations at a distribution margin, southern California. Journal of Biogeography. 38(9):1708 1716. doi:10.1111/j.1365-2699.2011.02525.x.

Harte J. 2011. Maximum Entropy and Ecology: A Theory of Abundance, Distribution, and Energetics. Oxford, UK: Oxford University Press.

He F, Gaston KJ. 2003. Occupancy, spatial variance, and the abundance of species. The American Naturalist. 162(3):366 375. doi:10.1086/377190.

Hedges LV, Gurevitch J, Curtis PS. 1999. The meta-analysis of response ratios in experimental ecology. Ecology. 80(4):1150 1156. doi:10.1890/0012-9658(1999)080[1150:TMAORR]2.0.CO;2.

Hickling R, Roy DB, Hill JK, Fox R, Thomas CD. 2006. The distributions of a wide range of taxonomic groups are expanding polewards. Global Change Biology. 12(3):450 455. doi:10.1111/j.1365-2486.2006.01116.x.

Hitch AT, Leberg PL. 2007. Breeding distributions of North American bird species moving north as a result of climate change. Conservation Biology. 21(2):534 539. doi:10.1111/j.1523-1739.2006.00609.x.

64

Holtrop AM, Cao Y, Dolan CR. 2010. Estimating sampling effort required for characterizing species richness and site-to-site similarity in fish assemblage surveys of wadeable Illinois streams. Transactions of the American Fisheries Society. 139(5):1421 1435. doi:10.1577/T09-078.1.

Holyoak M, Leibold MA, Holt RD, editors. 2005. Metacommunities: Spatial Dynamics and Ecological Communities. Chicago, IL: University of Chicago Press.

Hooten MB, Hobbs NT. 2015. A guide to Bayesian model selection for ecologists. Ecological Monographs. 85(1):3 28. doi:10.1890/14-0661.1.

Hovick TJ, Allred BW, McGranahan DA, Palmer MW, Elmore RD, Fuhlendorf SD. 2016. Informing conservation by identifying range shift patterns across breeding habitats and migration strategies. Biodiversity and Conservation. 25(2):345 356. doi:10.1007/s10531-016-1053-6.

Huang Q, Sauer JR, Dubayah RO. 2017. Multidirectional abundance shifts among North American birds and the relative influence of multifaceted climate factors. Global Change Biology. 23(9):3610 3622. doi:10.1111/gcb.13683.

Hubbell SP. 2001. The Unified Neutral Theory of Biodiversity and Biogeography. Princeton, NJ: Princeton University Press.

Ims RA, Henden JA. 2012. Collapse of an arctic bird community resulting from ungulate-induced loss of erect shrubs. Biological Conservation. 149(1):2 5. doi:10.1016/j.biocon.2012.02.008.

Jensen A, Meilby H. 2012. Assessing the population status of a tree species using distance sampling: Aquilaria crassna (Thymelaeaceae) in Northern Laos. International Journal of Forestry Research. 2012(265831):1 11. doi:10.1155/2012/265831.

Jetz W, Thomas GH, Joy JB, Redding DW, Hartmann K, Mooers AO. 2014. Global distribution and conservation of evolutionary distinctness in birds. Current Biology. 24(9):919 930. doi:10.1016/j.cub.2014.03.011.

Joseph LN, Elkin C, Martin TG, Possingham HP. 2009. Modeling abundance using N-mixture models: The importance of considering ecological mechanisms. Ecological Applications. 19(3):631 642. doi:10.1890/07-2107.1.

Karanth KK, Nichols JD, Sauer JR, Hines JE. 2006. Comparative dynamics of avian communities across edges and interiors of North American ecoregions. Journal of Biogeography. 33(4):674 682. doi:10.1111/j.1365-2699.2005.01392.x.

65

Kark S, Rensburg BJ van. 2006. Ecotones: Marginal or central areas of transition? Israel Journal of Ecology & Evolution. 52(1):29 53. doi:10.1560/IJEE.52.1.29.

Keith SA, Newton AC, Morecroft MD, Golicher DJ, Bullock JM. 2011. Plant metacommunity structure remains unchanged during biodiversity loss in English woodlands. Oikos. 120(2):302 310. doi:10.1111/j.1600-0706.2010.18775.x.

Kelt DA. 1999. On the relative importance of history and ecology in structuring communities of desert small animals. Ecography. 22(2):123 137. doi:10.1111/j.1600-0587.1999.tb00460.x.

Kéry M. 2008. Estimating abundance from bird counts: Binomial mixture models uncover complex covariate relationships. The Auk. 125(2):336 345. doi:10.1525/auk.2008.125.issue-2.

Kéry M, Royle JA. 2008. Hierarchical Bayes estimation of species richness and occupancy in spatially replicated surveys. Journal of Applied Ecology. 45(2):589 598. doi:10.1111/j.1365-2664.2007.01441.x.

Kéry M, Royle JA. 2009. Inference about species richness and community structure using species-specific occupancy models in the national Swiss breeding bird survey MHB. In: Thomson DL, Cooch EG, Conroy MJ, editors. Modeling Demographic Processes in Marked Populations. New York, NY: Springer. p. 639 656.

Kéry M, Royle JA, Plattner M, Dorazio RM. 2009. Species richness and occupancy estimation in communities subject to temporary emigration. Ecology. 90(5):1279 1290. doi:10.1890/07-1794.1.

Kéry M, Royle JA, Schmid H. 2005. Modeling avian abundance from replicated counts using binomial mixture models. Ecological Applications. 15(4):1450 1461. doi:10.1890/04-1120.

Kéry M, Schaub M. 2012. Bayesian Population Analysis using WinBUGS: A Hierarchical Perspective. Boston, MA: Academic Press.

727. doi:10.2307/1934919.

Koleff P, Gaston KJ, Lennon JJ. 2003. Measuring beta diversity for presence absence data. Journal of Animal Ecology. 72(3):367 382. doi:10.1046/j.1365-2656.2003.00710.x.

Kujala H, Vepsäläinen V, Zuckerberg B, Brommer JE. 2012. Range margin shifts of birds revisited the role of spatiotemporally varying survey effort. Global Change Biology. 19(2):420 430. doi:10.1111/gcb.12042.

66

La Sorte FA, Thompson FR. 2007. Poleward Shifts in Winter Ranges of North American Birds. Ecology. 88(7):1803 1812. doi:10.1890/06-1072.1.

Lane LJ, Romney EM, Hakonson TE. 1984. Water balance calculations and net production of perennial vegetation in the northern Mojave Desert. Journal of Range Management. 37(1):12 18. doi:10.2307/3898814.

Lehikoinen A, Virkkala R. 2015. North by north-west: Climate change and directions of density shifts in birds. Global Change Biology. 22(3):1121 1129. doi:10.1111/gcb.13150.

Leibold MA, Holyoak M, Mouquet N, Amarasekare P, Chase JM, Hoopes MF, Holt RD, Shurin JB, Law R, Tilman D, et al. 2004. The metacommunity concept: A framework for multi-scale community ecology. Ecology Letters. 7(7):601 613. doi:10.1111/j.1461-0248.2004.00608.x.

Leibold MA, Mikkelson GM. 2002. Coherence, species turnover, and boundary clumping: Elements of meta-community structure. Oikos. 97(2):237 250. doi:10.1034/j.1600-0706.2002.970210.x.

Levin SA. 1992. The problem of pattern and scale in ecology: The Robert H. MacArthur award lecture. Ecology. 73(6):1943 1967. doi:10.2307/1941447.

Link WA, Sauer JR. 1996. Extremes in ecology: Avoiding the misleading effects of sampling variation in summary analyses. Ecology. 77(5):1633 1640. doi:10.2307/2265557.

Lunn D, Spiegelhalter D, Thomas A, Best N. 2009. The BUGS project: Evolution, critique and future directions. Statistics in Medicine. 28(25):3049 3067. doi:10.1002/sim.3680.

Lunn DJ, Thomas A, Best N, Spiegelhalter D. 2000. WinBUGS - A Bayesian modelling framework: Concepts, structure, and extensibility. Statistics and Computing. 10(4):325 337. doi:10.1023/A:1008929526011.

MacKenzie DI. 2018. Occupancy Estimation and Modeling: Inferring Patterns and Dynamics of Species Occurrence. 2nd ed. London, UK: Academic Press.

MacKenzie DI, Nichols JD, Hines JE, Knutson MG, Franklin AB. 2003. Estimating site occupancy, colonization, and local extinction when a species is detected imperfectly. Ecology. 84(8):2200 2207. doi:10.1890/02-3090.

MacKenzie DI, Nichols JD, Lachman GB, Droege S, Andrew Royle J, Langtimm CA. 2002. Estimating site occupancy rates when detection probabilities are less than one. Ecology. 83(8):2248 2255. doi:10.1890/0012-9658(2002)083[2248:ESORWD]2.0.CO;2.

67

MacKenzie DI, Nichols JD, Royle JA, Pollock KH, Bailey LL, Hines JE. 2006. Occupancy Estimation and Modeling: Inferring Patterns and Dynamics of Species Occurrence. Burlington, MA: Academic Press.

MacLean SA, Rios Domingues A, de Valpine P, Beissinger SR. A century of climate and land-use

Global Change Biology. In review.

MacMahon JA. 1979. North American deserts: Their floral and faunal components. In: Goodall DW, Perry RA, editors. Arid Land Ecosystems: Structure, Functioning and Management. Vol. 1. London, UK: Cambridge University Press. p. 21 82.

MacNeil MA, Tyler EHM, Fonnesbeck CJ, Rushton SP, Polunin NVC, Conroy MJ. 2008. Accounting for detectability in reef-fish biodiversity estimates. Marine Ecology Progress Series. 367:249 260. doi:10.3354/meps07580.

Magurran AE. 2004. Measuring Biological Diversity. Oxford, UK: Blackwell.

Mao CX, Colwell RK. 2005. Estimation of species richness: Mixture models, the role of rare species, and inferential challenges. Ecology. 86(5):1143 1153. doi:10.1890/04-1078.

Margules CR, Pressey RL. 2000. Systematic conservation planning. Nature. 405(6783):243 253. doi:10.1038/35012251.

Martin J, Royle JA, Mackenzie DI, Edwards HH, Kéry M, Gardner B. 2011. Accounting for non-independent detection when estimating abundance of organisms with a Bayesian approach. Methods in Ecology and Evolution. 2(6):595 601. doi:10.1111/j.2041-210X.2011.00113.x.

Massimino D, Johnston A, Pearce-Higgins JW. 2015. The geographical range of British birds expands during 15 years of warming. Bird Study. 62(4):523 534. doi:10.1080/00063657.2015.1089835.

Mattsson BJ, Zipkin EF, Gardner B, Blank PJ, Sauer JR, Royle JA. 2013. Explaining local-scale species distributions: Relative contributions of spatial autocorrelation and landscape heterogeneity for an avian assemblage. PLoS ONE. 8(2):e55097. doi:10.1371/journal.pone.0055097.

Maurer BA, Brown JH. 1989. Distributional consequences of spatial variation in local demographic processes. Annales Zoologici Fennici. 26(2):121 131. https://www.jstor.org/stable/23736063.

68

McClintock BT, Bailey LL, Pollock KH, Simons TR. 2010. Experimental investigation of observation error in anuran call surveys. The Journal of Wildlife Management. 74(8):18821893. doi:10.2193/2009-321.

McCreedy C, van Riper C. 2014. Drought-caused delay in nesting of Sonoran Desert birds and its facilitation of parasite- and predator-mediated variation in reproductive success. The Auk. 132(1):235 247. doi:10.1642/AUK-13-253.1.

Mehlman DW. 1997. Change in avian abundance across the geographic range in response to environmental change. Ecological Applications. 7(2):614 624. doi:10.1890/1051-0761(1997)007[0614:CIAAAT]2.0.CO;2.

Mihaljevic JR, Joseph MB, Johnson PTJ. 2015. Using multispecies occupancy models to improve the characterization and understanding of metacommunity structure. Ecology. 96(7):17831792. doi:10.1890/14-1580.1.

Millar RB. 2009. Comparison of hierarchical Bayesian models for overdispersed count data using 969. doi:10.1111/j.1541-0420.2008.01162.x.

Miller AH. 1951. An Analysis of the Distribution of the Birds of California. Berkeley, CA: University of California Press.

Miller DA, Nichols JD, McClintock BT, Grant EHC, Bailey LL, Weir LA. 2011. Improving occupancy estimation when two types of observational error occur: Non-detection and species misidentification. Ecology. 92(7):1422 1428. doi:10.1890/10-1396.1.

Moilanen A, Wilson KA, Possingham HP. 2009. Spatial Conservation Prioritization: Quantitative Methods and Computational Tools. Oxford, UK: Oxford University Press.

Mokany K, Ferrier S. 2011. Predicting impacts of climate change on biodiversity: A role for semi-mechanistic community-level modelling. Diversity and Distributions. 17(2):374 380. doi:10.1111/j.1472-4642.2010.00735.x.

Moreno-Rueda G. 2010. European bird species have expanded northwards during 1950-1993 in response to recent climatic warming. In: Ulrich PK, Willett JH, editors. Trends in Ornithology Research. Hauppauge, NY: Nova Science Publishers, Inc. p. 137 151.

Morisita M. 1962. II -Index, a measure of dispersion of individuals. Researches on Population Ecology. 4(1):1 7. doi:10.1007/BF02533903.

69

Moritz C, Patton JL, Conroy CJ, Parra JL, White GC, Beissinger SR. 2008. Impact of a century of climate change on small-mammal communities in Yosemite National Park, USA. Science. 322(5899):261 264. doi:10.1126/science.1163428.

Morand predator-mediated processes. Oecologia. 133(3):315 324. doi:10.1007/s00442-002-1040-3.

Moses MR, Frey JK, Roemer GW. 2012. Elevated surface temperature depresses survival of banner-tailed kangaroo rats: Will climate change cook a desert icon? Oecologia. 168(1):257268. doi:10.1007/s00442-011-2073-2.

Myhrvold NP, Baldridge E, Chan B, Sivam D, Freeman DL, Ernest SKM. 2015. An amniote life-history database to perform comparative analyses with birds, mammals, and reptiles. Ecology. 96(11):3109 3109. doi:10.1890/15-0846R.1.

Newton AC, Walls RM, Golicher D, Keith SA, Diaz A, Bullock JM. 2011. Structure, composition and dynamics of a calcareous grassland metacommunity over a 70-year interval. Journal of Ecology. 100(1):196 209. doi:10.1111/j.1365-2745.2011.01923.x.

Nice MM. 1941. The role of territory in bird life. The American Midland Naturalist. 26(3):441487. doi:10.2307/2420732.

North American Bird Conservation Initiative, U. S. Committee. 2014. The State of the Birds 2014 Report. Washington, DC: US Department of Interior. [accessed 2018 Jan 1]. http://www.stateofthebirds.org/2014/.

how and which. Bayesian Analysis. 4(1):85 117. doi:10.1214/09-BA403.

Simpson GL, Solymos P, et al. 2018. vegan: Community Ecology Package.

Ovaskainen O, Soininen J. 2010. Making more out of sparse data: Hierarchical modeling of species communities. Ecology. 92(2):289 295. doi:10.1890/10-1251.1.

Parmesan C, Yohe G. 2003. A globally coherent fingerprint of climate change impacts across natural systems. Nature. 421(6918):37 42. doi:10.1038/nature01286.

Partners in Flight. 2017. Avian Conservation Assessment Database, version 2017. [accessed 2017 Nov 5]. http://pif.birdconservancy.org/ACAD.

70

Patten DT, Rouse L, Stromberg JC. 2008. Isolated spring wetlands in the Great Basin and Mojave Deserts, USA: Potential response of vegetation to groundwater withdrawal. Environmental Management. 41(3):398 413. doi:10.1007/s00267-007-9035-9.

Pattinson NB, Smit B. 2017. Seasonal behavioral responses of an arid-zone passerine in a hot environment. Physiology & Behavior. 179(Supplement C):268 275. doi:10.1016/j.physbeh.2017.06.018.

Paxton EH, Camp RJ, Gorresen PM, Crampton LH, Leonard DL, VanderWerf EA. 2016. Collapsing avian community on a Hawaiian island. Science Advances. 2(9):e1600029. doi:10.1126/sciadv.1600029.

Pearce-Higgins JW, Eglington SM, Martay B, Chamberlain DE. 2015. Drivers of climate change impacts on bird communities. Journal of Animal Ecology. 84(4):943 954. doi:10.1111/1365-2656.12364.

Peel MC, Finlayson BL, Mcmahon TA. 2007. Updated world map of the Köppen-Geiger climate classification. Hydrology and Earth System Sciences Discussions. 4(2):439 473. doi:10.5194/hess-11-1633-2007.

Peters DPC, Gosz JR, Pockman WT, Small EE, Parmenter RR, Collins SL, Muldavin E. 2006. Integrating patch and boundary dynamics to understand and predict biotic transitions at multiple scales. Landscape Ecology. 21(1):19 33. doi:10.1007/s10980-005-1063-3.

du Plessis KL, Martin RO, Hockey PAR, Cunningham SJ, Ridley AR. 2012. The costs of keeping cool in a warming world: Implications of high temperatures for foraging, thermoregulation and body condition of an arid-zone bird. Global Change Biology. 18(10):3063 3070. doi:10.1111/j.1365-2486.2012.02778.x.

Plummer M. 2003. JAGS: A program for analysis of Bayesian graphical models using Gibbs sampling. In: Hornik K, Leisch F, Zeileis A, editors. Proceedings of the 3rd International Workshop on Distributed Statistical Computing (DSC 2003). Vol. 124. Vienna, AT. p. 2022.

Post E. 2013. Ecology of Climate Change: The Importance of Biotic Interactions. Princeton, NJ: Princeton University Press.

Presley SJ, Higgins CL, Willig MR. 2010. A comprehensive framework for the evaluation of metacommunity structure. Oikos. 119(6):908 917. doi:10.1111/j.1600-0706.2010.18544.x.

Preston FW. 1948. The commonness, and rarity, of species. Ecology. 29(3):254 283. doi:10.2307/1930989.

71

Preston KL, Rotenberry JT, Redak RA, Allen MF. 2008. Habitat shifts of endangered species under altered climate conditions: Importance of biotic interactions. Global Change Biology. 14(11):2501 2515. doi:10.1111/j.1365-2486.2008.01671.x.

Price SJ, Garner TWJ, Nichols RA, Balloux F, Ayres C, Mora-Cabello de Alba A, Bosch J. 2014. Collapse of amphibian communities due to an introduced Ranavirus. Current Biology. 24(21):2586 2591. doi:10.1016/j.cub.2014.09.028.

R Core Team. 2017. R: A language and environment for statistical computing. Version 3.4.0. Vienna, AT: R Foundation for Statistical Computing. [accessed 2017 Apr 30]. http://www.R-project.org.

Ralph CJ, Droege S, Sauer JR. 1995. Managing and monitoring birds using point counts: Standards and applications. In: Ralph CJ, Droege S, Sauer JR, editors. Monitoring Bird Populations by Point Counts. Albany, CA: Pacific Southwest Research Station, Forest Service, U.S. Department of Agriculture. p. 161 169.

Ralston J, DeLuca WV, Feldman RE, King DI. 2017. Population trends influence species ability to track climate change. Global Change Biology. 23(4):1390 1399. doi:10.1111/gcb.13478.

Randall JM, Parker SS, Moore J, Cohen B, Crane L, Christian B, Cameron D, MacKenzie JB, Klausmeyer K, Morrison S. 2010. Mojave Desert Ecoregional Assessment. The Nature Conservancy. [accessed 2018 Jan 1]. https://www.scienceforconservation.org/products/mojave-desert-ecoregional-assessment.

Rapacciuolo G, Maher SP, Schneider AC, Hammond TT, Jabis MD, Walsh RE, Iknayan KJ, Walden GK, Oldfather MF, Ackerly DD, et al. 2014. Beyond a warming fingerprint: Individualistic biogeographic responses to heterogeneous climate change in California. Global Change Biology. 20(9):2841 2855. doi:10.1111/gcb.12638.

Rehfeldt GE, Crookston NL, Sáenz-Romero C, Campbell EM. 2012. North American vegetation model for land-use planning in a changing climate: A solution to large classification problems. Ecological Applications. 22(1):119 141. doi:10.1890/11-0495.1.

Richmond OMW, Hines JE, Beissinger SR. 2010. Two-species occupancy models: A new parameterization applied to co-occurrence of secretive rails. Ecological Applications. 20(7):2036 2046. doi:10.1890/09-0470.1.

Rodewald P. 2015. The Birds of North America. Ithaca, NY: Cornell Laboratory of Ornithology. [accessed 2017 Jul 8]. https://birdsna.org.

72

Rodríguez JP. 2002. Range contraction in declining North American bird populations. Ecological Applications. 12(1):238 248. doi:10.1890/1051-0761(2002)012[0238:RCIDNA]2.0.CO;2.

Rota CT, Fletcher Jr RJ, Dorazio RM, Betts MG. 2009. Occupancy estimation and the closure assumption. Journal of Applied Ecology. 46(6):1173 1181. doi:10.1111/j.1365-2664.2009.01734.x.

Rowe KC, Rowe KMC, Tingley MW, Koo MS, Patton JL, Conroy CJ, Perrine JD, Beissinger SR, Moritz C. 2015. Spatially heterogeneous impact of climate change on small mammals of montane California. Proceedings of the Royal Society of London B: Biological Sciences. 282(1799):20141857. doi:10.1098/rspb.2014.1857.

Royle JA. 2004. N-Mixture models for estimating population size from spatially replicated counts. Biometrics. 60(1):108 115. doi:10.1111/j.0006-341X.2004.00142.x.

Royle JA, Dorazio RM. 2008. Hierarchical Modeling and Inference in Ecology: The Analysis of Data from Populations, Metapopulations and Communities. 1st ed. Boston, MA: Academic.

Royle JA, Dorazio RM. 2012. Parameter-expanded data augmentation for Bayesian analysis of capture recapture models. Journal of Ornithology. 152(2):521 537. doi:10.1007/s10336-010-0619-4.

Royle JA, Kéry M. 2007. A Bayesian state-space formulation of dynamic occupancy models. Ecology. 88(7):1813 1823. doi:10.1890/06-0669.1.

Royle JA, Link WA. 2006. Generalized site occupancy models allowing for false positive and false negative errors. Ecology. 87(4):835 841. doi:10.1890/0012-9658(2006)87[835:GSOMAF]2.0.CO;2.

Royle JA, Nichols JD. 2003. Estimating abundance from repeated presence absence data or point counts. Ecology. 84(3):777 790. doi:10.1890/0012-9658(2003)084[0777:EAFRPA]2.0.CO;2.

Ruiz-Gutiérrez V, Zipkin EF, Dhondt AA. 2010. Occupancy dynamics in a tropical bird community: Unexpectedly high forest use by birds classified as non-forest species. Journal of Applied Ecology. 47(3):621 630. doi:10.1111/j.1365-2664.2010.01811.x.

Russell RE, Royle JA, Saab VA, Lehmkuhl JF, Block WM, Sauer JR. 2009. Modeling the effects of environmental disturbance on wildlife communities: Avian responses to prescribed fire. Ecological Applications. 19(5):1253 1263. doi:10.1890/08-0910.1.

73

Sagarin RD, Gaines SD, Gaylord B. 2006. Moving beyond assumptions to understand abundance distributions across the ranges of species. Trends in Ecology & Evolution. 21(9):524 530. doi:10.1016/j.tree.2006.06.008.

Sauer JR, Pardieck KL, Ziolkowski DJ, Smith AC, Hudson M-AR, Rodriguez V, Berlanga H, Niven DK, Link WA. 2017. The first 50 years of the North American Breeding Bird Survey. The Condor. 119(3):576 593. doi:10.1650/CONDOR-17-83.1.

Shannon C. 1948. A mathematical theory of communication. Bell System Technical Journal. 27(3):379 423. doi:10.1002/j.1538-7305.1948.tb01338.x.

Shefferson RP, Proper J, Beissinger SR, Simms EL. 2003. Life history trade-offs in a rare orchid: The costs of flowering, dormancy, and sprouting. Ecology. 84(5):1199 1206. doi:10.1890/0012-9658(2003)084[1199:LHTIAR]2.0.CO;2.

Shmida A, Wilson MV. 1985. Biological determinants of species diversity. Journal of Biogeography. 12(1):1 20. doi:10.2307/2845026.

Simons TR, Alldredge MW, Pollock KH, Wettroth JM, Dufty Jr AM. 2007. Experimental analysis of the auditory detection process on avian point counts. The Auk. 124(3):986 999. doi:10.1642/0004-8038(2007)124[986:EAOTAD]2.0.CO;2.

Simpson EH. 1949. Measurement of diversity. Nature. 163(4148):688 688. doi:10.1038/163688a0.

Simpson GG. 1943. Mammals and the nature of continents. American Journal of Science. 241(1):1 31. doi:10.2475/ajs.241.1.1.

Sinervo B, Méndez-de-la-Cruz F, Miles DB, Heulin B, Bastiaans E, Cruz MVS, Lara-Resendiz R, Martínez-Méndez N, Calderón-Espinosa ML, Meza-Lázaro RN, et al. 2010. Erosion of lizard diversity by climate change and altered thermal niches. Science. 328(5980):894 899. doi:10.1126/science.1184695.

Small A. 1994. California Birds: Their Status and Distribution. Vista, CA: Ibis Pub. Co.

Smit B, McKechnie AE. 2015. Water and energy fluxes during summer in an arid-zone passerine bird. Ibis. 157(4):774 786. doi:10.1111/ibi.12284.

Smit B, Zietsman G, Martin RO, Cunningham SJ, McKechnie AE, Hockey PAR. 2016. Behavioural responses to heat in desert birds: Implications for predicting vulnerability to climate warming. Climate Change Responses. 3:9. doi:10.1186/s40665-016-0023-2.

74

Socolar JB, Epanchin PN, Beissinger SR, Tingley MW. 2017. Phenological shifts conserve thermal niches in North American birds and reshape expectations for climate-driven range shifts. Proceedings of the National Academy of Sciences. 114(49):12976 12981. doi:10.1073/pnas.1705897114.

Sólymos P, Lele S, Bayne E. 2012. Conditional likelihood approach for analyzing single visit abundance survey data in the presence of zero inflation and detection error. Environmetrics. 23(2):197 205. doi:10.1002/env.1149.

Sørensen T. 1948. A method of establishing groups of equal amplitude in plant sociology based on similarity of species and its application to analyses of the vegetation on Danish commons. Royal Danish Academy of Sciences and Letters. 5:1 34.

Spiegelhalter DJ, Best NG, Carlin BP, Van Der Linde A. 2002. Bayesian measures of model complexity and fit. Journal of the Royal Statistical Society: Series B (Statistical Methodology). 64(4):583 639. doi:10.1111/1467-9868.00353.

Svenning JC, Sandel B. 2013. Disequilibrium vegetation dynamics under future climate change. American Journal of Botany. 100(7):1266 1286. doi:10.3732/ajb.1200469.

Tang G, Arnone JA. 2013. Trends in surface air temperature and temperature extremes in the Great Basin during the 20th century from ground-based observations. Journal of Geophysical Research: Atmospheres. 118(9):3579 3589. doi:10.1002/jgrd.50360.

Tayleur C, Caplat P, Massimino D, Johnston A, Jonzén N, Smith HG, Lindström Å. 2015. Swedish birds are tracking temperature but not rainfall: Evidence from a decade of abundance changes. Global Ecology and Biogeography. 24(7):859 872. doi:10.1111/geb.12308.

Thomas CD, Lennon JJ. 1999. Birds extend their ranges northwards. Nature. 399(6733):213. doi:10.1038/20335.

Tingley MW, Beissinger SR. 2013. Cryptic loss of montane avian richness and high community turnover over 100 years. Ecology. 94(3):598 609. doi:10.1890/12-0928.1.

Tingley MW, Koo MS, Moritz C, Rush AC, Beissinger SR. 2012. The push and pull of climate change causes heterogeneous shifts in avian elevational ranges. Global Change Biology. 18(11):3279 3290. doi:10.1111/j.1365-2486.2012.02784.x.

Tingley MW, Monahan WB, Beissinger SR, Moritz C. 2009. Birds track their Grinnellian niche through a century of climate change. Proceedings of the National Academy of Sciences of

75

the United States of America. 106(Supplement 2):19637 19643. doi:10.1073/pnas.0901562106.

Tinkham ER. 1948. Faunistic and ecological studies on the Orthoptera of the Big Bend region of Trans-Pecos Texas, with especial reference to the Orthopteran zones and faunae of midwestern North America. The American Midland Naturalist. 40(3):521 663. doi:10.2307/2421489.

Udvardy MDF. 1963. Bird faunas of North America. In: Proceedings of the XIII International

1147 1167.

Vanderwal J, Murphy HT, Kutt AS, Perkins GC, Bateman BL, Perry JJ, Reside AE. 2013. Focus

Nature Climate Change. 3(3):239 243. doi:10.1038/nclimate1688.

Virkkala R, Lehikoinen A. 2014. Patterns of climate-induced density shifts of species: Poleward shifts faster in northern boreal birds than in southern birds. Global Change Biology. 20(10):2995 3003. doi:10.1111/gcb.12573.

Waddle JH, Dorazio RM, Walls SC, Rice KG, Beauchamp J, Schuman MJ, Mazzotti FJ. 2010. A new parameterization for estimating co-occurrence of interacting species. Ecological Applications. 20(5):1467 1475. doi:10.1890/09-0850.1.

Walls SC, Waddle JH, Dorazio RM. 2011. Estimating occupancy dynamics in an anuran assemblage from Louisiana, USA. Journal of Wildlife Management. 75(4):751 761. doi:10.1002/jwmg.97.

Walther GR. 2010. Community and ecosystem responses to recent climate change. Philosophical Transactions of the Royal Society B: Biological Sciences. 365(1549):2019 2024. doi:10.1098/rstb.2010.0021.

Weiss JL, Castro CL, Overpeck JT. 2009. Distinguishing pronounced droughts in the Southwestern United States: Seasonality and effects of warmer temperatures. J Climate. 22(22):5918 5932. doi:10.1175/2009JCLI2905.1.

Weiss JL, Overpeck JT. 2005. Is the Sonoran Desert losing its cool? Global Change Biology. 11(12):2065 2077. doi:10.1111/j.1365-2486.2005.01020.x.

Wenger SJ, Freeman MC. 2008. Estimating species occurrence, abundance, and detection probability using zero-inflated distributions. Ecology. 89(10):2953 2959. doi:10.1890/07-1127.1.

76

Westerling AL, Hidalgo HG, Cayan DR, Swetnam TW. 2006. Warming and earlier spring increase western U.S. forest wildfire activity. Science. 313(5789):940 943. doi:10.1126/science.1128834.

Whelan CJ, Wenny DG, Marquis RJ. 2008. Ecosystem services provided by birds. Annals of the New York Academy of Sciences. 1134(1):25 60. doi:10.1196/annals.1439.003.

Whittaker RH. 1965. Dominance and diversity in land plant communities: Numerical relations of species express the importance of competition in community function and evolution. Science. 147(3655):250 260. doi:10.1126/science.147.3655.250.

Wieczorek J, Guo Q, Hijmans R. 2004. The point-radius method for georeferencing locality descriptions and calculating associated uncertainty. International Journal of Geographical Information Science. 18(8):745 767. doi:10.1080/13658810412331280211.

Williams AP, Seager R, Abatzoglou JT, Cook BI, Smerdon JE, Cook ER. 2015. Contribution of anthropogenic warming to California drought during 2012 2014. Geophysical Research Letters. 42(16):2015GL064924. doi:10.1002/2015GL064924.

Williams JB, Tieleman BI. 2005. Physiological adaptation in desert birds. BioScience. 55(5):416425. doi:10.1641/0006-3568(2005)055[0416:PAIDB]2.0.CO;2.

Williams MR. 2009. Butterflies and day-flying moths in a fragmented urban landscape, south-west Western Australia: Patterns of species richness. Pacific Conservation Biology. 15(1):3246. doi:10.1071/PC090032.

Wilson JB. 1991. Methods for fitting dominance/diversity curves. Journal of Vegetation Science. 2(1):35 46. doi:10.2307/3235896.

Wolf B. 2000. Global warming and avian occupancy of hot deserts; a physiological and behavioral perspective. Revista Chilena De Historia Natural. 73(3):395 400. doi:10.4067/s0716-078x2000000300003.

Wu J, Shi Y. 2016. Attribution index for changes in migratory bird distributions: The role of climate change over the past 50 years in China. Ecological Informatics. 31:147 155. doi:10.1016/j.ecoinf.2015.11.013.

Wu J, Zhang G. 2015. Can changes in the distributions of resident birds in China over the past 50 years be attributed to climate change? Ecology and Evolution. 5(11):2215 2233. doi:10.1002/ece3.1513.

77

Wuebbles DJ, Fahey DW, Hibbard KA, Dokken DJ, Stewart BC, Maycock TK. 2017. Climate Science Special Report: Fourth National Climate Assessment, Volume I. Washington, DC: US Global Change Research Program.

Xue T, Tang G, Sun L, Wu Y, Liu Y, Dou Y. 2017. Long-term trends in precipitation and precipitation extremes and underlying mechanisms in the U.S. Great Basin during 19512013. Journal of Geophysical Research: Atmospheres. 122(12):6152 6169. doi:10.1002/2017JD026682.

Yamaura Y, Royle J, Shimada N, Asanuma S, Sato T, Taki H, Makino S. 2012. Biodiversity of man-made open habitats in an underused country: A class of multispecies abundance models for count data. Biodiversity and Conservation. 21(6):1365 1380. doi:10.1007/s10531-012-0244-z.

Zhou L, Chen H, Dai Y. 2015. Stronger warming amplification over drier ecoregions observed since 1979. Environmental Research Letters. 10(6):064012. doi:10.1088/1748-9326/10/6/064012.

Zipkin EF, Andrew Royle J, Dawson DK, Bates S. 2010. Multi-species occurrence models to evaluate the effects of conservation and management actions. Biological Conservation. 143(2):479 484. doi:10.1016/j.biocon.2009.11.016.

Zipkin EF, DeWan A, Andrew Royle J. 2009. Impacts of forest fragmentation on species richness: A hierarchical approach to community modelling. Journal of Applied Ecology. 46(4):815822. doi:10.1111/j.1365-2664.2009.01664.x.

Zipkin EF, Grant EHC, Fagan WF. 2012. Evaluating the predictive abilities of community occupancy models using AUC while accounting for imperfect detection. Ecological Applications. 22(7):1962 1972. doi:10.1890/11-1936.1.

Zuckerberg B, Woods AM, Porter WF. 2009. Poleward shifts in breeding bird distributions in New York State. Global Change Biology. 15(8):1866 1883. doi:10.1111/j.1365-2486.2009.01878.x.

78

Appendix A Appendix Tables

Table A.1 Species Traits Included in Relative Occupancy Change Comparison. Abbreviations are included for traits, territoriality [A: A-type, O: other; as defined by(1941)], elevational preference (H: high, L: low), range margin (N: northern, S: southern, P: patchy, F: range full encompassed study area), habitat (F: forest, A: arid lands, F: forest, Gen.: generalist), diet (H: herbivore, O: omnivore, I: insectivore, C: carnivore), and migratory status (R: residential, T: temperate, N: Neotropical). Traits were compiled from the following sources: mi-gratory mode, elevational preference, range ma The Birds of North America Online (Rodewald 2015); habitat specialization from Partners in Flight (2017); diet breadth from De Graaf et al. (1985); and clutch size and body mass from Myhrvold et al. (2015). Range margins were categorized by the presence of a range boundary within the study area, as defined by expert-created distribution maps (Rodewald 2015).

taxonomic treatment. Reclassification of the recently-quivocal, thus were aggregated as the historic species complex the Sage Sparrow.

Common name Scientific name Family Clut

ch si

ze

Body

mas

s (g)

Terr

itori

ality

Elev

. pre

f.

Rang

e mar

gin

Hab

itat

Die

t

Mig

. sta

tus

Mountain Quail Oreortyx pictus Odontophoridae 10.0 233.0 A H P F H R California Quail Callipepla californica Odontophoridae 14.0 171.5 O H N A H R

Callipepla gambelii Odontophoridae 12.8 169.2 O L P A H R Chukar Alectoris chukar Phasianidae 11.2 535.5 A L S A H R Wild Turkey Meleagris gallopavo Phasianidae 11.0 5811.0 O H P F O R Eurasian Collared-Dove Streptopelia decaocto Columbidae 2.0 154.7 A L F Gn H R White-winged Dove Zenaida asiatica Columbidae 2.0 153.0 O L N A H T Mourning Dove Zenaida macroura Columbidae 2.0 119.0 O L F Gn H R Greater Roadrunner Geococcyx californianus Cuculidae 4.0 376.0 A L N A C R White-throated Swift Aeronautes saxatalis Apodidae 4.3 31.9 O L S A I R Blk-chd Hummingbird Archilochus alexandri Trochilidae 2.0 3.4 A H S F O N

Calypte anna Trochilidae 2.0 4.4 A H P A O R Calypte costae Trochilidae 2.0 3.1 A L P A O T

Bd-tailed Hummingbird Selasphorus platycercus Trochilidae 2.0 3.5 A H S F O N Killdeer Charadrius vociferus Charadriidae 4.0 94.3 A L F Gn I R Turkey Vulture Cathartes aura Cathartidae 2.0 1776.0 O L F Gn C N Sharp-shinned Hawk Accipiter striatus Accipitridae 4.2 138.5 A L S F C R

Accipiter cooperii Accipitridae 4.3 452.0 O L F F C R Buteo swainsoni Accipitridae 2.5 969.6 O L S G C N

Red-tailed Hawk Buteo jamaicensis Accipitridae 2.8 1126.0 A L F Gn C R Golden Eagle Aquila chrysaetos Accipitridae 2.0 4383.0 A H F Gn C R Burrowing Owl Athene cunicularia Strigidae 8.0 150.6 O L F G C R Red-naped Sapsucker Sphyrapicus nuchalis Picidae 4.9 50.3 A H P F O T Ld-backed Woodpecker Picoides scalaris Picidae 4.0 30.6 A L P A O R Hairy Woodpecker Picoides villosus Picidae 3.9 65.9 A H S F O R Northern Flicker Colaptes auratus Picidae 6.6 130.7 O H S F I R American Kestrel Falco sparverius Falconidae 4.6 115.5 A L F Gn C R Prairie Falcon Falco mexicanus Falconidae 4.2 754.5 O L F A C R Olive-sided Flycatcher Contopus cooperi Tyrannidae 3.0 32.8 A H P F I N

79

Common name Scientific name Family Clut

ch si

ze

Body

mas

s (g)

Terr

itori

ality

Elev

. pre

f.

Rang

e mar

gin

Hab

itat

Die

t

Mig

. sta

tus

Western Wood-Pewee Contopus sordidulus Tyrannidae 3.0 12.8 A H P F I N Willow Flycatcher Empidonax traillii Tyrannidae 3.5 13.2 A L P F I N Gray Flycatcher Empidonax wrightii Tyrannidae 3.5 12.4 A H S F I N Dusky Flycatcher Empidonax oberholseri Tyrannidae 3.6 10.4 A H P F I N Pacific-slope Flycatcher Empidonax difficilis Tyrannidae 3.3 10.8 A H N F I N Cordilleran Flycatcher Empidonax occidentalis Tyrannidae 3.3 11.0 A H P F I N Black Phoebe Sayornis nigricans Tyrannidae 4.0 18.2 A H N G I R

Sayornis saya Tyrannidae 4.5 21.0 A L F A I R Vermilion Flycatcher Pyrocephalus rubinus Tyrannidae 2.9 14.0 A L F F I R Ash-throated Flycatcher Myiarchus cinerascens Tyrannidae 4.4 28.2 A L F A I N Western Kingbird Tyrannus verticalis Tyrannidae 3.8 39.6 O L P G I N Loggerhead Shrike Lanius ludovicianus Laniidae 5.0 47.2 A L F G C R

Vireo bellii Vireonidae 3.9 8.6 A L F A I N Gray Vireo Vireo vicinior Vireonidae 3.9 12.8 A H P F I T

reo Vireo cassinii Vireonidae 3.9 15.3 A H N F I N Plumbeous Vireo Vireo plumbeus Vireonidae 3.9 16.0 A H P F I N Warbling Vireo Vireo gilvus Vireonidae 3.6 13.1 A H P F I N Pinyon Jay Gymnorhinus

cyanocephalus Corvidae 3.8 105.0 O H P F O R

Jay Cyanocitta stelleri Corvidae 4.0 120.0 O H P F O R California Scrub-Jay Aphelocoma californica Corvidae 4.0 82.8 A H N F O R

-Jay Aphelocoma woodhouseii Corvidae 4.0 82.8 A H S F O R Nucifraga columbiana Corvidae 3.0 130.0 O H P F O R

Common Raven Corvus corax Corvidae 4.8 1097.0 A L F Gn O R Horned Lark Eremophila alpestris Alaudidae 3.5 34.2 A L F G O R Violet-green Swallow Tachycineta thalassina Hirundinidae 4.5 14.2 O H F F I N N R-winged Swallow Stelgidopteryx serripennis Hirundinidae 6.0 15.8 O L F Gn I N Cliff Swallow Petrochelidon pyrrhonota Hirundinidae 4.2 22.4 O L S Gn I N Barn Swallow Hirundo rustica Hirundinidae 4.5 19.0 O H S Gn I N Mountain Chickadee Poecile gambeli Paridae 7.0 11.5 A H S F I R Oak Titmouse Baeolophus inornatus Paridae 6.8 16.3 A H N F O R Juniper Titmouse Baeolophus ridgwayi Paridae 6.9 16.4 A H S F O R Verdin Auriparus flaviceps Remizidae 3.5 7.0 A L N A I R Bushtit Psaltriparus minimus Aegithalidae 6.0 5.3 O H P F I R Wh-breasted Nuthatch Sitta carolinensis Sittidae 8.0 21.0 A H S F I R Pygmy Nuthatch Sitta pygmaea Sittidae 7.0 10.6 A H P F O R Brown Creeper Certhia americana Certhiidae 5.5 9.0 A H P F I R Rock Wren Salpinctes obsoletus Troglodytidae 5.2 16.5 A L F A I R Canyon Wren Catherpes mexicanus Troglodytidae 5.5 11.6 A L F A I R House Wren Troglodytes aedon Troglodytidae 6.3 11.0 A H N F I R Marsh Wren Cistothorus palustris Troglodytidae 5.0 11.2 A L P G I R

Thryomanes bewickii Troglodytidae 5.8 9.9 A H P A I R Cactus Wren Campylorhynchus

brunneicapillus Troglodytidae 3.9 38.9 A L P A O R

Blue-gray Gnatcatcher Polioptila caerulea Polioptilidae 4.2 6.1 A H F Gn I T Black-tailed Gnatcatcher Polioptila melanura Polioptilidae 4.0 5.2 A L N A I R Gdn-crowned Kinglet Regulus satrapa Regulidae 8.0 6.1 A H N F I R

80

Common name Scientific name Family Clut

ch si

ze

Body

mas

s (g)

Terr

itori

ality

Elev

. pre

f.

Rang

e mar

gin

Hab

itat

Die

t

Mig

. sta

tus

Ruby-crowned Kinglet Regulus calendula Regulidae 8.0 6.6 A H P F I R Wrentit Chamaea fasciata Sylviidae 3.7 14.7 A L P A I R Western Bluebird Sialia mexicana Turdidae 5.0 27.1 O L P F I R Mountain Bluebird Sialia currucoides Turdidae 5.2 29.6 A H S F I R

Myadestes townsendi Turdidae 3.8 33.5 A H S F O R Catharus ustulatus Turdidae 3.6 30.1 A H N F O N

Hermit Thrush Catharus guttatus Turdidae 3.7 30.6 A H S F I T American Robin Turdus migratorius Turdidae 3.4 78.8 A H S F O T

Toxostoma bendirei Mimidae 3.5 62.2 A L P A O T California Thrasher Toxostoma redivivum Mimidae 3.0 84.4 A H N A O R

Toxostoma lecontei Mimidae 3.1 61.9 A L N A I R Crissal Thrasher Toxostoma crissale Mimidae 2.5 62.7 A H N A O R Sage Thrasher Oreoscoptes montanus Mimidae 4.1 44.2 A H S A I T Northern Mockingbird Mimus polyglottos Mimidae 3.9 48.5 A L F Gn O R European Starling Sturnus vulgaris Sturnidae 5.0 77.7 O L F Gn O R Phainopepla Phainopepla nitens Ptiliogonatidae 2.5 22.6 A L N A H T House Sparrow Passer domesticus Passeridae 4.3 27.7 O L F Gn H R House Finch Haemorhous mexicanus Fringillidae 4.3 21.4 O L F Gn H R Cassii Haemorhous cassinii Fringillidae 4.5 27.5 O H P F O R Red Crossbill Loxia curvirostra Fringillidae 3.4 39.5 O H P F O R Pine Siskin Spinus pinus Fringillidae 3.5 13.1 O H P F O R Lesser Goldfinch Spinus psaltria Fringillidae 4.0 9.1 O L F F H R

Spinus lawrencei Fringillidae 4.5 10.8 O H N A H T Green-tailed Towhee Pipilo chlorurus Passerellidae 4.0 29.3 A H S A O T Spotted Towhee Pipilo maculatus Passerellidae 3.6 39.9 A H S F O R Rf-crowned Sparrow Aimophila ruficeps Passerellidae 3.7 18.8 A H P A O R California Towhee Melozone crissalis Passerellidae 3.2 47.0 A H N A O R Chipping Sparrow Spizella passerina Passerellidae 4.0 12.3 A H P F O T

Spizella breweri Passerellidae 3.1 10.9 A H P A I T Black-chinned Sparrow Spizella atrogularis Passerellidae 3.0 11.6 A H P A O T Vesper Sparrow Pooecetes gramineus Passerellidae 4.0 25.1 A L P G O T Lark Sparrow Chondestes grammacus Passerellidae 4.0 28.5 O L S G O T Black-throated Sparrow Amphispiza bilineata Passerellidae 3.0 13.5 A L F A I T Sage Sparrow Artemisiospiza belli Passerellidae 3.4 18.1 A L P A O R Fox Sparrow Passerella iliaca Passerellidae 4.0 33.8 A H N F O T Song Sparrow Melospiza melodia Passerellidae 4.0 23.4 A H P Gn O R White-crowned Sparrow Zonotrichia leucophrys Passerellidae 4.0 28.0 A H N F O R Dark-eyed Junco Junco hyemalis Passerellidae 4.0 19.5 A H P F O R Yellow-breasted Chat Icteria virens Icteriidae 3.8 25.3 A L N F O N Yellow-headed Blackbird

Xanthocephalus xanthocephalus

Icteridae 3.7 72.5 O L P F O T

Western Meadowlark Sturnella neglecta Icteridae 4.8 97.7 A L F G I R Hooded Oriole Icterus cucullatus Icteridae 4.0 24.3 O L N F O N

Icterus bullockii Icteridae 4.5 35.2 O L F F O N Icterus parisorum Icteridae 3.0 36.4 O L P F O T

Red-winged Blackbird Agelaius phoeniceus Icteridae 3.3 54.0 A L F Gn O R Brown-headed Cowbird Molothrus ater Icteridae 4.3 42.5 O L F Gn O T

81

Common name Scientific name Family Clut

ch si

ze

Body

mas

s (g)

Terr

itori

ality

Elev

. pre

f.

Rang

e mar

gin

Hab

itat

Die

t

Mig

. sta

tus

Euphagus cyanocephalus Icteridae 5.0 62.6 O L F Gn O T Great-tailed Grackle Quiscalus mexicanus Icteridae 3.3 168.7 A L N Gn O R Or-crowned Warbler Oreothlypis celata Parulidae 4.5 9.2 A H P F I T

Oreothlypis luciae Parulidae 4.5 6.6 A L N A I N Oreothlypis virginiae Parulidae 3.8 8.2 A H S F I N

Geothlypis tolmiei Parulidae 4.0 10.5 A H P F I N Common Yellowthroat Geothlypis trichas Parulidae 4.0 9.7 A L F Gn I T Yellow Warbler Setophaga petechia Parulidae 4.4 10.0 A L S F I N

r Setophaga graciae Parulidae 3.1 8.1 A H S F I N Blk-thted Gray Warbler Setophaga nigrescens Parulidae 4.0 8.2 A H S F I N Hermit Warbler Setophaga occidentalis Parulidae 4.0 10.3 A H N F O N Western Tanager Piranga ludoviciana Cardinalidae 3.8 30.0 A H P F O N Black-headed Grosbeak Pheucticus

melanocephalus Cardinalidae 3.4 45.5 A H S F O N

Blue Grosbeak Passerina caerulea Cardinalidae 3.8 28.4 A L N F O N Lazuli Bunting Passerina amoena Cardinalidae 4.0 15.2 A H S F O N

82

Table A.2 Relationship of Modeled Probabilities to Covariates Summary of community- and species-level coefficient values from a global dynamic multispecies occupancy model. Coefficients whose 95% credible interval do not cross zero, are bolded. The count of each species with species-level coefficients non-overlapping with zero are summarized by the direction, positive or negative, of the response.

Probability Covariate Mean 95% CI Negative Positive Initial occupancy Elevation (m) -0.541 -1.162, 0.068 26 4 Elevation, quadratic -0.475 * -0.689, -0.275 24 5 Historic precipitation (mm) 0.339 * 0.066, 0.609 3 15 Historic mean temperature

(°C) -0.44 -1.017, 0.143 0 0

Detection Julian day 0.264 * 0.084, 0.455 8 24 Julian day, quadratic -0.286 * -0.425, -0.155 23 2 Time period -1.716 * -1.933, -1.501 119 0 Persistence Years elapsed -0.202 -0.584, 0.211 0 0 Colonization Years elapsed 0.371 * 0.061, 0.696 0 1

83

Input for JAGS model

Data and Model Variables det_matrix: encounter histories (1 if a species was encountered on a visit, 0 if not encountered)

organized in a 4-dimensional array by species, visit, site, and time period det_covs: detection covariates specific to a survey visit at a site. The covariates included in this

analysis are Julian day and its quadratic. Organized in a 4-dimensional array by visit, site, time period, and covariate

site_det_covs: detection covariates specific to a site during a time period. The covariate in-cluded in this analysis is a categorical variable of whether the survey occurred in the historic time period (1) or the modern time period (0). Organized in a 3-dimensional array site, time period, and covariate.

colon_covs: colonization covariates specific to a site for time periods beyond the initial time step.The covariate included in this analysis is years elapsed between the historic and modern surveys. Organized in a 3-dimensional array site, time period, and covariate.

surv_covs: survival (persistence) covariates specific to a site for time periods beyond the initial time step. Organized in a 3-dimensional array site, time period, and covariate. See main text for covariates included in the model.

occ_covs: intial occupancy covariates specific to a site during the historic time period. Covari-ates included in this analysis: historic mean temperature, historic annual precipitation, ele-vation and elevations quadratic.

Likelihood loop indexes: n_species: 135 avian species visits_by_site: a matrix of the total number of visits at a site in each time period (61 x 2) n_sites: 61 sites n_time_periods: 2 time periods n_occ_covs: 4 covariates n_det_covs: 2 detection covariates + 1 site detection covariate n_colon_covs: 1 covariate n_surv_covs: 11 covariates n_site_det_covs: 1 covariate

Note on Priors The priors for the community-level standard deviations for the intercept and coefficients are spec-ified as coming from a half-Cauchy distribution. The Cauchy distribution is called in JAGS using the Student t distribution with degrees of freedom set to one (dt(0, tau, 1)). Censoring the dis-tribution above zero (T(0, )) creates the half-Cauchy. All standard deviations are transformed to precisions using pow(x, -2)

84

JAGS Model Code

model{ # Community-level (hyper) priors -------------------------------------------- tau <- pow(2.25, -2) # # Intercepts: # Intial occupancy occ_int_location ~ dnorm(0, tau) occ_int_sigma ~ dt(0, tau, 1) T(0, ) occ_int_scale <- pow(occ_int_sigma, -2) # Detection det_int_location ~ dnorm(0, tau) det_int_sigma ~ dt(0, tau, 1) T(0, ) det_int_scale <- pow(det_int_sigma, -2) # Survival and colonization surv_int_sigma ~ dt(0, tau, 1) T(0, ) colon_int_sigma ~ dt(0, tau, 1) T(0, ) for (time in 1:(n_time_periods - 1)) { surv_int_location[time] ~ dnorm(0, tau) surv_int_scale[time] <- pow(surv_int_sigma, -2) colon_int_location[time] ~ dnorm(0, tau) colon_int_scale[time] <- pow(colon_int_sigma, -2) } # # Coefficients: # Initial occupancy for (cov in 1:n_occ_covs) { occ_coef_sigma[cov] ~ dt(0, tau, 1) T(0, ) occ_coef_location[cov] ~ dnorm(0, tau) occ_coef_scale[cov] <- pow(occ_coef_sigma[cov], -2) } # Detection for (cov in 1:n_det_covs) { det_coef_sigma[cov] ~ dt(0, tau, 1) T(0, ) det_coef_location[cov] ~ dnorm(0, tau) det_coef_scale[cov] <- pow(det_coef_sigma[cov], -2) } # Survival and colonization for (cov in 1:n_surv_covs) { surv_coef_sigma[cov] ~ dt(0, tau, 1) T(0, ) } for (cov in 1:n_colon_covs) { colon_coef_sigma[cov] ~ dt(0, tau, 1) T(0, ) } for (time in 1:(n_time_periods - 1)) { for (cov in 1:n_surv_covs) { surv_coef_location[time, cov] ~ dnorm(0, tau) surv_coef_scale[time, cov] <- pow(surv_coef_sigma[cov], -2) } for (cov in 1:n_colon_covs) { colon_coef_location[time, cov] ~ dnorm(0, tau) colon_coef_scale[time, cov] <- pow(colon_coef_sigma[cov], -2) } } # Likelihood loop ----------------------------------------------------------- for (species in 1:n_species) {

85

# # Species-level intercept and coefficient priors: # Occupancy occ_int[species] ~ dnorm(occ_int_location, occ_int_scale) for (cov in 1:n_occ_covs) { occ_coef[species, cov] ~ dnorm(occ_coef_location[cov], occ_coef_scale[cov]) } # Survival and colonization for (time in 1:(n_time_periods - 1)) { surv_int[species, time] ~ dnorm(surv_int_location[time], surv_int_scale[time]) for (cov in 1:n_surv_covs) { surv_coef[species, time, cov] ~ dnorm(surv_coef_location[time, cov], surv_coef_scale[time, cov]) } colon_int[species, time] ~ dnorm(colon_int_location[time], colon_int_scale[time]) for (cov in 1:n_colon_covs) { colon_coef[species, time, cov] ~ dnorm(colon_coef_location[time, cov], colon_coef_scale[time, cov]) } } for (site in 1:n_sites) { # # Regressions: # Initial occupancy state (time = 1) logit(occ[species, site, 1]) <- occ_int[species] + inprod(occ_coef[species, ], occ_covs[site, ]) mean_incidence[species, site, 1] <- occ[species, site, 1] # Model of changes in occupancy state for time=2, ..., n for (time in 1:(n_time_periods - 1)) { logit(surv[species, site, time]) <- surv_int[species, time] + inprod(surv_coef[species, time, ], surv_covs[site, time, ]) logit(colon[species, site, time]) <- colon_int[species, time] + inprod(colon_coef[species, time, ], colon_covs[site, time, ]) mean_incidence[species, site, time + 1] <- surv[species, site, time] * incidence[species, site, time] + colon[species, site, time] * (1 - incidence[species, site, time]) } } # # Detection loops: # Species-level intercept and coefficient priors det_int[species] ~ dnorm(det_int_location, det_int_scale) for (cov in 1:n_det_covs) { det_coef[species, cov] ~ dnorm(det_coef_location[cov], det_coef_scale[cov]) } for (site in 1:n_sites) { for (time in 1:n_time_periods) { for (visit in 1:visits_by_site[site, time]) { # Regression index <- n_det_covs - n_site_det_covs site_index <- (index + 1):n_det_covs logit(det[species, visit, site, time]) <- det_int[species] + inprod(det_coef[species, 1:index], det_covs[visit, site, time, ]) + inprod(det_coef[species, site_index], site_det_covs[site, time, ])

86

mean_det_matrix[species, visit, site, time] <- det[species, visit, site, time] * incidence[species, site, time] # Generate a likelihood for the MCMC sampler by linking the # model to the field data det_matrix[species, visit, site, time] ~ dbern(mean_det_matrix[species, visit, site, time]) # Simulate the detection matrix that is predicted by the model # (used to calculate PPC) det_matrix_simulated[species, visit, site, time] ~ dbern(mean_det_matrix[species, visit, site, time]) } # Partially-latent incidence matrix incidence[species, site, time] ~ dbern(mean_incidence[species, site, time]) } } } }

Post-processing of MCMC Output

Integrated likelihood The integrated likelihood is a multispecies adaptation of Mackenzie et al. (MacKenzie et al. 2003). The probability of an occupancy state in the first time period for a species ( ) and a site ( ) (occu-pied = , unoccupied = ) is represented by:

(A.1)

The following matrix summarizes all possible tranistions in occupancy states from time period to . The rows represent the occupancy state in time . The columns represent the occupancy

state in time . If a site is occupied in time the species can either persist at a site with proba-bility or become locally extinct with probability . If a site is unoccupied in time , the species either colonizes the site with probability , or the site remains uncolonized with proba-bility .

(A.2)

The probability of observing the encounter history for species across all visits at site in sampling period is given by the following equation where is the detection probability of a species at a site during a visit ( ). The notation of the first entry of the matrix behaves as an

-then- statement, depending on the whether or not the species was observed during a par-ticular visit to site during time period . If a species is observed during visit (i.e., ), the probability of the encounter history during that visit is ; if the species is unobserved (i.e.,

), the probability of the detection history is . The overall probability of an en-counter history at a site is the product of the visit-level probabilities. The second entry of the matrix

87

is an indicator function, which has the value of 1 if a species has an all-zero encounter history, and is 0 if the species is observed at least once at a site across all visits.

(A.3)

The likelihood of observing the entire set of encounter histories for a species across all time periods at a site is calculated as a combination of the results of equations A.1-A.3:

(A.4)

R Code to Calculate Likelihood

calculate_likelihood <- function(eh, probs, simulated = F) { n_iterations <- dim(probs$detection)[1] n_periods <- dim(probs$detection)[5] # Instead of using matrix multiplication for each species, sites and # iteration, which would more closely follow the notation of MacKenzie et al # 2003. Breaking the calculations into element-wise components is more # performant. # Dimensions of probability data # detection: [iteration, species, visit, site, time_period] # all others: [iteration, species, site, time_period] # Dimensions of encounter history change depending on wheter it is # simulated or not. The simulate data is returned from JAGS as # det_matrix_simulated and is used to calculate the PPC/Bayesian p-value. # simulated: [iteration, species, visit, site, time_period] # field data: [species, visit, site, time_period] # Step 1: create the encounter-history probability matrix ------------------ # Create the encounter-history probability matrix (p_Xijt). Equivalent to # equation 3. # Element one: combined detection probs across all visits at a site # (during a time period) # Element two: indicator = 1 if encounter history at that site is all zero # For each visit, return the detection probability if observered and # 1 - detection probability if unobserved calculate_combined_detection <- function(detection, eh){ combined_detection <- abs(1 - eh - detection) } # na.rm = T handles imbalanced-survey design. if (simulated == T) { probs$detection <- abs(1 - eh - probs$detection)

88

# NB: colSums/aperm is faster than apply/sum non_det_indicator <- colSums(aperm(eh, c(3, 1, 2, 4, 5)), na.rm = T) } else { # sweep() applies the field data to the data estimated by the model # for each iteration. probs$detection <- sweep(probs$detection, c(2, 3, 4, 5), eh, FUN = calculate_combined_detection) non_det_indicator <- colSums(aperm(eh, c(2, 1, 3, 4)), na.rm = T) } non_det_indicator <- ifelse(non_det_indicator == 0, 1, 0) # If real data, replicate non_det_indicator for number of iterations if (simulated == F) { non_det_indicator <- replicate(n_iterations, non_det_indicator) non_det_indicator <- aperm(non_det_indicator, c(4, 1, 2, 3)) } # Take product of incidence-modified detection probabilities at the site # NB: array_prod is faster than using apply/prod array_prod <- function(x){ x[is.na(x)] <- 1 x_update <- x[ , , 1, , ] for(i in 2:dim(x)[3]){ x_update <- x_update * x[ , , i, , ] } return(x_update) } combined_detection <- array_prod(probs$detection) # Step 2: combine time-step probs ------------------------------------------ # NB: steps 2 - 5 surmise equation 4 # multiply the encounter-history probability matrix (p_Xijt) with the # transition probability matrix (delta_ijt), excluding p_Xijt of the final # time period. combine_transition <- function(detection_element, probability_element){ element <- detection_element[ , , , 1:(n_periods - 1), drop = F] * probability_element } detection_survival <- combine_transition(combined_detection, probs$survival) detection_extinction <- combine_transition(combined_detection, (1 - probs$survival)) indicator_colonization <- combine_transition(non_det_indicator, probs$colonization) indicator_noncolonization <- combine_transition(non_det_indicator, (1 - probs$colonization)) # Step 3: multiply the time-step probs ------------------------------------- # Take each matrix element, make time slice # Initialize with first time step result_ds <- detection_survival[ , , , 1] result_de <- detection_extinction[ , , , 1] result_ic <- indicator_colonization[ , , , 1]

89

result_inc <- indicator_noncolonization[ , , , 1] if (n_periods > 2) { for(t in 2:(n_periods - 1)){ temp_ds <- result_ds * detection_survival[ , , , t] + result_de * indicator_colonization[ , , , t] temp_de <- result_ds * detection_extinction[ , , , t] + result_de * indicator_noncolonization[ , , , t] temp_ic <- result_ic * detection_survival[ , , , t] + result_inc * indicator_colonization[ , , , t] temp_inc <- result_ic * detection_extinction[ , , , t] + result_inc * indicator_noncolonization[ , , , t] # Set result for the next time step result_ds <- temp_ds result_de <- temp_de result_ic <- temp_ic result_inc <- temp_inc } } # Step 4: multiply transitions with final time step ----------------------- combined_survival <- result_ds * combined_detection[ , , , n_periods] + result_de * non_det_indicator[ , , , n_periods] combined_extinction <- result_ic * combined_detection[ , , , n_periods] + result_inc * non_det_indicator[ , , , n_periods] # Step 5: multiply transitions with initial occupancy ---------------------- site_likelihood <- probs$occupancy[ , , , 1] * combined_survival + (1 - probs$occupancy[ , , , 1]) * combined_extinction # Step 6: combine site-level likelihoods ----------------------------------- # Equation 5 likelihood <- apply(site_level_likelihood, c(1, 2), prod) return(likelihood) }

Posterior Predictive Check (Bayesian P-value) We used a goodness-of-fit test statistic based on deviance residuals following Broms et al. (2016). The posterior predictive check was calculated from the comparison of deviances of the observed data and that simulated by the model. The deviance of the observed data was calculated using the integrated likelihood (see calculation above) for each iteration of the MCMC sample:

(A.5)

and was compared to the deviance of the detections simulated from the model ( ):

(A.6)

The posterior predictive check is calculated as the proportion of times the deviance of the observed data is greater than the deviance of the predicted.

90

(A.6)

An ideal fit is a value of 0.5 indicating the deviance of the observed data is larger than the predicted data half of the time; values beyond the extremes of 0.05 and 0.95 is considered an inadequate fit (Gelman et al. 1996; Broms et al. 2016).

Calculation of PPC in R

calculate_bayesian_p <- function(model_likelihood, simulated_likelihood) { n_iterations <- dim(model_likelihood)[1] # Calculate deviance for each iteration observed_deviance <- -2 * rowSums(log(model_likelihood)) simulated_deviance <- -2 * rowSums(log(simulated_likelihood)) bayesian_p <- sum(ifelse(observed_deviance > simulated_deviance, 1, 0)) bayesian_p <- bayesian_p / n_iterations return(bayesian_p) }

91

Appendix B JAGS model

Data and Model Variables det_matrix: encounter histories (1 if a species was encountered on a visit, 0 if not encountered)

organized in a 4-dimensional array by species, visit, site, and time period det_covs: detection covariates specific to a survey visit at a site. The covariates included in this

analysis are Julian day and its quadratic. Organized in a 4-dimensional array by visit, site, time period, and covariate

site_det_covs: detection covariates specific to a site during a time period. The covariate in-cluded in this analysis is a categorical variable of whether the survey occurred in the historic time period (1) or the modern time period (0). Organized in a 3-dimensional array site, time period, and covariate.

mod_covs: historic occupancy covariates specific to a site during the historic time period. Co-variates included in this analysis: latitude and latitude quadratic.

mod_covs: modern occupancy covariates specific to a site during the modern time period. Co-variates included in this analysis: latitude and latitude quadratic.

Likelihood loop indexes: n_species: 162 avian species visits_by_site: a matrix of the total number of visits at a site in each time period (106 x 2) n_sites: 106 sites n_time_periods: 2 time periods n_hist_covs: 2 covariates for historic occupancy probability n_mod_covs: 2 covariates for modern occupancy probability n_det_covs: 2 detection covariates + 1 site detection covariate n_site_det_covs: 1 covariate

Note on Priors The priors for the community-level standard deviations for the intercept and coefficients are spec-ified as coming from a half-Cauchy distribution. The Cauchy distribution is called in JAGS using the Student t distribution with degrees of freedom set to one (dt(0, tau, 1)). Censoring the dis-tribution above 0 (T(0, )) creates the half-Cauchy. All standard deviations are transformed to precisions using pow(x, -2)

92

JAGS Model Code

model { tau <- 1 / (2.25 ^ 2) # Hyper- & sp-level priors ----------------------------------------------------- for (cov in 1:n_hist_covs) { hist_mu[cov] ~ dnorm(0, tau) hist_tau[cov] ~ dt(0, tau, 1) T(0, ) for(sp in 1:n_species){ hist_coefs[sp, cov] ~ dnorm(hist_mu[cov], hist_tau[cov]) } } for (cov in 1:n_mod_covs) { mod_mu[cov] ~ dnorm(0, tau) mod_tau[cov] ~ dt(0, tau, 1) T(0, ) for(sp in 1:n_species){ mod_coefs[sp, cov] ~ dnorm(mod_mu[cov], mod_tau[cov]) } } for (cov in 1:n_det_covs) { det_mu[cov] ~ dnorm(0, tau) det_tau[cov] ~ dt(0, tau, 1) T(0, ) for(sp in 1:n_species){ det_coefs[sp, cov] ~ dnorm(det_mu[cov], det_tau[cov]) } } auto_mu ~ dnorm(0, tau) auto_tau ~ dt(0, tau, 1) T(0, ) # Likelihood loop ----------------------------------------------------------- for (sp in 1:n_species) { auto[sp] ~ dnorm(auto_mu, auto_tau) for (st in 1:n_sites) { # Initial occupancy state (time = 1) logit(occ[sp, st, 1]) <- hist_coefs[sp, 1] + inprod(hist_coefs[sp, 2:n_hist_covs], hist_covs[st, ]) # Occupancy state for time = 2 logit(occ[sp, st, 2]) <- mod_coefs[sp, 1] + inprod(mod_coefs[sp, 2:n_mod_covs], mod_covs[st, ]) + auto[sp] * incidence[sp, st, 1] for (tm in 1:n_time_periods) { # Partially-latent incidence matrix incidence[sp, st, tm] ~ dbern(occ[sp, st, tm]) for (vt in 1:visits_by_site[st, tm]){ # # Detection: logit(det[sp, vt, st, tm]) <- det_coefs[sp, 1] + inprod(det_coefs[sp, visit_index], visit_det_covs[vt, st, tm, ]) + inprod(det_coefs[sp, site_index], site_det_covs[st, tm, ]) mu_det_matrix[sp, vt, st, tm] <- det[sp, vt, st, tm] * incidence[sp, st, tm] # Generate a likelihood for the MCMC sampler by linking the # model to the field data det_matrix[sp, vt, st, tm] ~ dbern(mu_det_matrix[sp, vt, st, tm]) } }

93

} } }