Reframing wildlife disease management problems with decision analysis.

Contemporary wildlife disease management is complex because managers need to respond to a wide range of stakeholders, multiple uncertainties, and difficult trade-offs that characterize the interconnected challenges of today. Despite general acknowledgment of these complexities, managing wildlife disease tends to be framed as a scientific problem, in which the major challenge is lack of knowledge. The complex and multifactorial process of decision-making is collapsed into a scientific endeavor to reduce uncertainty. As a result, contemporary decision-making may be oversimplified, rely on simple heuristics, and fail to account for the broader legal, social, and economic context in which the decisions are made. Concurrently, scientific research on wildlife disease may be distant from this decision context, resulting in information that may not be directly relevant to the pertinent management questions. We propose reframing wildlife disease management challenges as decision problems and addressing them with decision analytical tools to divide the complex problems into more cognitively manageable elements. In particular, structured decision-making has the potential to improve the quality, rigor, and transparency of decisions about wildlife disease in a variety of systems. Examples of management of severe acute respiratory syndrome coronavirus 2, white-nose syndrome, avian influenza, and chytridiomycosis illustrate the most common impediments to decision-making, including competing objectives, risks, prediction uncertainty, and limited resources.

Replanteamiento del manejo de problemas por enfermedades de fauna mediante el análisis de decisiones Resumen El manejo actual de las enfermedades de la fauna es complejo debido a que los gestores necesitan responder a una amplia gama de actores, varias incertidumbres y compensaciones difíciles que caracterizan los retos interconectados del día de hoy. A pesar de que en general se reconocen estas complejidades, el manejo de las enfermedades tiende a plantearse como un problema científico en el que el principal obstáculo es la falta de conocimiento. El proceso complejo y multifactorial de la toma decisiones está colapsado dentro de un esfuerzo científico para reducir la incertidumbre. Como resultado de esto, las decisiones contemporáneas pueden estar simplificadas en exceso, depender de métodos heurísticos simples y no considerar el contexto legal, social y económico más amplio en el que se toman las decisiones. De manera paralela, las investigaciones científicas sobre las enfermedades de la fauna pueden estar lejos de este contexto de decisiones, lo que deriva en información que puede no ser directamente relevante para las preguntas pertinentes de manejo. Proponemos replantear los obstáculos para el manejo de enfermedades de fauna como problemas de decisión y abordarlos con herramientas analíticas de decisión para dividir los problemas complejos en elementos más manejables de manera cognitiva. En particular, las decisiones estructuradas tienen el potencial de mejorar la calidad, el rigor y la transparencia de las decisiones sobre las enfermedades de la fauna en una variedad de sistemas. Ejemplos como el manejo del coronavirus del síndrome de respiración agudo tipo 2, el síndrome de nariz blanca, la influenza aviar y la quitridiomicosis ilustran los impedimentos más comunes para la toma de decisiones, incluyendo los objetivos en competencia, riesgos, incertidumbre en las predicciones y recursos limitados.

Broad-Scale Assessment of Methylmercury in Adult Amphibians.

Mercury (Hg) is a toxic contaminant that has been mobilized and distributed worldwide and is a threat to many wildlife species. Amphibians are facing unprecedented global declines due to many threats including contaminants. While the biphasic life history of many amphibians creates a potential nexus for methylmercury (MeHg) exposure in aquatic habitats and subsequent health effects, the broad-scale distribution of MeHg exposure in amphibians remains unknown. We used nonlethal sampling to assess MeHg bioaccumulation in 3,241 juvenile and adult amphibians during 2017-2021. We sampled 26 populations (14 species) across 11 states in the United States, including several imperiled species that could not have been sampled by traditional lethal methods. We examined whether life history traits of species and whether the concentration of total mercury in sediment or dragonflies could be used as indicators of MeHg bioaccumulation in amphibians. Methylmercury contamination was widespread, with a 33-fold difference in concentrations across sites. Variation among years and clustered subsites was less than variation across sites. Life history characteristics such as size, sex, and whether the amphibian was a frog, toad, newt, or other salamander were the factors most strongly associated with bioaccumulation. Total Hg in dragonflies was a reliable indicator of bioaccumulation of MeHg in amphibians (R2 ≥ 0.67), whereas total Hg in sediment was not (R2 ≤ 0.04). Our study, the largest broad-scale assessment of MeHg bioaccumulation in amphibians, highlights methodological advances that allow for nonlethal sampling of rare species and reveals immense variation among species, life histories, and sites. Our findings can help identify sensitive populations and provide environmentally relevant concentrations for future studies to better quantify the potential threats of MeHg to amphibians.

Experimental evaluation of spatial capture-recapture study design.

A principal challenge impeding strong inference in analyses of wild populations is the lack of robust and long-term data sets. Recent advancements in analytical tools used in wildlife science may increase our ability to integrate smaller data sets and enhance the statistical power of population estimates. One such advancement, the development of spatial capture-recapture (SCR) methods, explicitly accounts for differences in spatial study designs, making it possible to equate multiple study designs in one analysis. SCR has been shown to be robust to variation in design as long as minimal sampling guidance is adhered to. However, these expectations are based on simulation and have yet to be evaluated in wild populations. Here we conduct a rigorously designed field experiment by manipulating the arrangement of artificial cover objects (ACOs) used to collect data on red-backed salamanders (Plethodon cinereus) to empirically evaluate the effects of design configuration on inference made using SCR. Our results suggest that, using SCR, estimates of space use and detectability are sensitive to study design configuration, namely the spacing and extent of the array, and that caution is warranted when assigning biological interpretation to these parameters. However, estimates of population density remain robust to design except when the configuration of detectors grossly violates existing recommendations.

A National-Scale Assessment of Mercury Bioaccumulation in United States National Parks Using Dragonfly Larvae As Biosentinels through a Citizen-Science Framework.

We conducted a national-scale assessment of mercury (Hg) bioaccumulation in aquatic ecosystems, using dragonfly larvae as biosentinels, by developing a citizen-science network to facilitate biological sampling. Implementing a carefully designed sampling methodology for citizen scientists, we developed an effective framework for a landscape-level inquiry that might otherwise be resource limited. We assessed the variation in dragonfly Hg concentrations across >450 sites spanning 100 United States National Park Service units and examined intrinsic and extrinsic factors associated with the variation in Hg concentrations. Mercury concentrations ranged between 10.4 and 1411 ng/g dry weight across sites and varied among habitat types. Dragonfly total Hg (THg) concentrations were up to 1.8-fold higher in lotic habitats than in lentic habitats and 37% higher in waterbodies with abundant wetlands along their margins than those without wetlands. Mercury concentrations in dragonflies differed among families but were correlated (r2 > 0.80) with each other, enabling adjustment to a consistent family to facilitate spatial comparisons among sampling units. Dragonfly THg concentrations were positively correlated with THg concentrations in both fish and amphibians from the same locations, indicating that dragonfly larvae are effective indicators of Hg bioavailability in aquatic food webs. We used these relationships to develop an integrated impairment index of Hg risk to aquatic ecosytems and found that 12% of site-years exceeded high or severe benchmarks of fish, wildlife, or human health risk. Collectively, this continental-scale study demonstrates the utility of dragonfly larvae for estimating the potential mercury risk to fish and wildlife in aquatic ecosystems and provides a framework for engaging citizen science as a component of landscape Hg monitoring programs.

Eco-evolutionary rescue promotes host-pathogen coexistence.

Emerging infectious pathogens are responsible for some of the most severe host mass mortality events in wild populations. Yet, effective pathogen control strategies are notoriously difficult to identify, in part because quantifying and forecasting pathogen spread and disease dynamics is challenging. Following an outbreak, hosts must cope with the presence of the pathogen, leading to host-pathogen coexistence or extirpation. Despite decades of research, little is known about host-pathogen coexistence post-outbreak when low host abundances and cryptic species make these interactions difficult to study. Using a novel disease-structured N-mixture model, we evaluate empirical support for three host-pathogen coexistence hypotheses (source-sink, eco-evolutionary rescue, and spatial variation in pathogen transmission) in a Neotropical amphibian community decimated by Batrachochytrium dendrobatidis (Bd) in 2004. During 2010-2014, we surveyed amphibians in Parque Nacional G. D. Omar Torríjos Herrera, Coclé Province, El Copé, Panama. We found that the primary driver of host-pathogen coexistence was eco-evolutionary rescue, as evidenced by similar amphibian survival and recruitment rates between infected and uninfected hosts. Average apparent monthly survival rates of uninfected and infected hosts were both close to 96%, and the expected number of uninfected and infected hosts recruited (via immigration/reproduction) was less than one host per disease state per 20-m site. The secondary driver of host-pathogen coexistence was spatial variation in pathogen transmission as we found that transmission was highest in areas of low abundance but there was no support for the source-sink hypothesis. Our results indicate that changes in the host community (i.e., through genetic or species composition) can reduce the impacts of emerging infectious disease post-outbreak. Our disease-structured N-mixture model represents a valuable advancement for conservation managers trying to understand underlying host-pathogen interactions and provides new opportunities to study disease dynamics in remnant host populations decimated by virulent pathogens.

Integrating count and detection-nondetection data to model population dynamics.

There is increasing need for methods that integrate multiple data types into a single analytical framework as the spatial and temporal scale of ecological research expands. Current work on this topic primarily focuses on combining capture-recapture data from marked individuals with other data types into integrated population models. Yet, studies of species distributions and trends often rely on data from unmarked individuals across broad scales where local abundance and environmental variables may vary. We present a modeling framework for integrating detection-nondetection and count data into a single analysis to estimate population dynamics, abundance, and individual detection probabilities during sampling. Our dynamic population model assumes that site-specific abundance can change over time according to survival of individuals and gains through reproduction and immigration. The observation process for each data type is modeled by assuming that every individual present at a site has an equal probability of being detected during sampling processes. We examine our modeling approach through a series of simulations illustrating the relative value of count vs. detection-nondetection data under a variety of parameter values and survey configurations. We also provide an empirical example of the model by combining long-term detection-nondetection data (1995-2014) with newly collected count data (2015-2016) from a growing population of Barred Owl (Strix varia) in the Pacific Northwest to examine the factors influencing population abundance over time. Our model provides a foundation for incorporating unmarked data within a single framework, even in cases where sampling processes yield different detection probabilities. This approach will be useful for survey design and to researchers interested in incorporating historical or citizen science data into analyses focused on understanding how demographic rates drive population abundance.

Modeling structured population dynamics using data from unmarked individuals.

The study of population dynamics requires unbiased, precise estimates of abundance and vital rates that account for the demographic structure inherent in all wildlife and plant populations. Traditionally, these estimates have only been available through approaches that rely on intensive mark-recapture data. We extended recently developed N-mixture models to demonstrate how demographic parameters and abundance can be estimated for structured populations using only stage-structured count data. Our modeling framework can be used to make reliable inferences on abundance as well as recruitment, immigration, stage-specific survival, and detection rates during sampling. We present a range of simulations to illustrate the data requirements, including the number of years and locations necessary for accurate and precise parameter estimates. We apply our modeling framework to a population of northern dusky salamanders (Desmognathus fuscus) in the mid-Atlantic region (USA) and find that the population is unexpectedly declining. Our approach represents a valuable advance in the estimation of population dynamics using multistate data from unmarked individuals and should additionally be useful in the development of integrated models that combine data from intensive (e.g., mark-recapture) and extensive (e.g., counts) data sources.

Updated range map of an endangered salamander and congeneric competitor reveals different niche preferences.

Estimating distributions for cryptic and highly range-restricted species induces unique challenges for species distribution modeling. In particular, bioclimatic covariates that are typically used to model species ranges at regional and continental scales may not show strong variation at scales of 100s and 10s of meters. This limits both the likelihood and usefulness of correlated occurrence to data typically used in distribution models. Here, we present analyses of species distributions, at 100 × 100 m resolution, for a highly range restricted salamander species (Shenandoah salamander, Plethodon shenandoah) and a closely related congener (red-backed salamander, Plethodon cinereus). We combined data across multiple survey types, account for seasonal variation in availability of our target species, and control for repeated surveys at locations- all typical challenges in range-scale monitoring datasets. We fit distribution models using generalized additive models that account for spatial covariates as well as unexplained spatial variation and spatial uncertainty. Our model accommodates different survey protocols using offsets and incorporates temporal variation in detection and availability resulting from survey-specific variation in temperature and precipitation. Our spatial random effect was crucial in identifying small-scale differences in the occurrence of each species and provides cell-specific estimates of uncertainty in the density of salamanders across the range. Counts of both species were seen to increase in the 3 days following a precipitation event. However, P. cinereus were observed even in extremely wet conditions, while surface activity of P. shenandoah was associated with a more narrow range. Our results demonstrate how a flexible analytical approach improves estimates of both distribution and uncertainty, and identify key abiotic relationships, even at small spatial scales and when scales of empirical data are mismatched. While our approach is especially valuable for species with small ranges, controlling for spatial autocorrelation, estimating spatial uncertainty, and incorporating survey-specific information in estimates can improve the reliability of distribution models in general.

A Dataset of Amphibian Species in U.S. National Parks.

National parks and other protected areas are important for preserving landscapes and biodiversity worldwide. An essential component of the mission of the United States (U.S.) National Park Service (NPS) requires understanding and maintaining accurate inventories of species on protected lands. We describe a new, national-scale synthesis of amphibian species occurrence in the NPS system. Many park units have a list of amphibian species observed within their borders compiled from various sources and available publicly through the NPSpecies platform. However, many of the observations in NPSpecies remain unverified and the lists are often outdated. We updated the amphibian dataset for each park unit by collating old and new park-level records and had them verified by regional experts. The new dataset contains occurrence records for 292 of the 424 NPS units and includes updated taxonomy, international and state conservation rankings, hyperlinks to a supporting reference for each record, specific notes, and related fields which can be used to better understand and manage amphibian biodiversity within a single park or group of parks.

Relaxing the closure assumption in occupancy models: staggered arrival and departure times.

Occupancy statistical models that account for imperfect detection have proved very useful in several areas of ecology, including species distribution and spatial dynamics, disease ecology, and ecological responses to climate change. These models are based on the collection of multiple samples at each of a number of sites within a given season, during which it is assumed the species is either absent or present and available for detection while each sample is taken. However, for some species, individuals are only present or available for detection seasonally. We present a statistical model that relaxes the closure assumption within a season by permitting staggered entry and exit times for the species of interest at each site. Based on simulation, our open model eliminates bias in occupancy estimators and in some cases increases precision. The power to detect the violation of closure is high if detection probability is reasonably high. In addition to providing more robust estimation of occupancy, this model permits comparison of phenology across sites, species, or years, by modeling variation in arrival or departure probabilities. In a comparison of four species of amphibians in Maryland we found that two toad species arrived at breeding sites later in the season than a salamander and frog species, and departed from sites earlier.

A strategy for monitoring and managing declines in an amphibian community.

Although many taxa have declined globally, conservation actions are inherently local. Ecosystems degrade even in protected areas, and maintaining natural systems in a desired condition may require active management. Implementing management decisions under uncertainty requires a logical and transparent process to identify objectives, develop management actions, formulate system models to link actions with objectives, monitor to reduce uncertainty and identify system state (i.e., resource condition), and determine an optimal management strategy. We applied one such structured decision-making approach that incorporates these critical elements to inform management of amphibian populations in a protected area managed by the U.S. National Park Service. Climate change is expected to affect amphibian occupancy of wetlands and to increase uncertainty in management decision making. We used the tools of structured decision making to identify short-term management solutions that incorporate our current understanding of the effect of climate change on amphibians, emphasizing how management can be undertaken even with incomplete information. Estrategia para Monitorear y Manejar Disminuciones en una Comunidad de Anfibios.

Evaluating the predictive abilities of community occupancy models using AUC while accounting for imperfect detection.

The ability to accurately predict patterns of species' occurrences is fundamental to the successful management of animal communities. To determine optimal management strategies, it is essential to understand species-habitat relationships and how species habitat use is related to natural or human-induced environmental changes. Using five years of monitoring data in the Chesapeake and Ohio Canal National Historical Park, Maryland, USA, we developed four multispecies hierarchical models for estimating amphibian wetland use that account for imperfect detection during sampling. The models were designed to determine which factors (wetland habitat characteristics, annual trend effects, spring/summer precipitation, and previous wetland occupancy) were most important for predicting future habitat use. We used the models to make predictions about species occurrences in sampled and unsampled wetlands and evaluated model projections using additional data. Using a Bayesian approach, we calculated a posterior distribution of receiver operating characteristic area under the curve (ROC AUC) values, which allowed us to explicitly quantify the uncertainty in the quality of our predictions and to account for false negatives in the evaluation data set. We found that wetland hydroperiod (the length of time that a wetland holds water), as well as the occurrence state in the prior year, were generally the most important factors in determining occupancy. The model with habitat-only covariates predicted species occurrences well; however, knowledge of wetland use in the previous year significantly improved predictive ability at the community level and for two of 12 species/species complexes. Our results demonstrate the utility of multispecies models for understanding which factors affect species habitat use of an entire community (of species) and provide an improved methodology using AUC that is helpful for quantifying the uncertainty in model predictions while explicitly accounting for detection biases.

Experimental investigation of false positive errors in auditory species occurrence surveys.

False positive errors are a significant component of many ecological data sets, which in combination with false negative errors, can lead to severe biases in conclusions about ecological systems. We present results of a field experiment where observers recorded observations for known combinations of electronically broadcast calling anurans under conditions mimicking field surveys to determine species occurrence. Our objectives were to characterize false positive error probabilities for auditory methods based on a large number of observers, to determine if targeted instruction could be used to reduce false positive error rates, and to establish useful predictors of among-observer and among-species differences in error rates. We recruited 31 observers, ranging in abilities from novice to expert, who recorded detections for 12 species during 180 calling trials (66,960 total observations). All observers made multiple false positive errors, and on average 8.1% of recorded detections in the experiment were false positive errors. Additional instruction had only minor effects on error rates. After instruction, false positive error probabilities decreased by 16% for treatment individuals compared to controls with broad confidence interval overlap of 0 (95% CI:--46 to 30%). This coincided with an increase in false negative errors due to the treatment (26%;--3 to 61%). Differences among observers in false positive and in false negative error rates were best predicted by scores from an online test and a self-assessment of observer ability completed prior to the field experiment. In contrast, years of experience conducting call surveys was a weak predictor of error rates. False positive errors were also more common for species that were played more frequently but were not related to the dominant spectral frequency of the call. Our results corroborate other work that demonstrates false positives are a significant component of species occurrence data collected by auditory methods. Instructing observers to only report detections they are completely certain are correct is not sufficient to eliminate errors. As a result, analytical methods that account for false positive errors will be needed, and independent testing of observer ability is a useful predictor for among-observer variation in observation error rates.

Improving occupancy estimation when two types of observational error occur: non-detection and species misidentification.

Efforts to draw inferences about species occurrence frequently account for false negatives, the common situation when individuals of a species are not detected even when a site is occupied. However, recent studies suggest the need to also deal with false positives, which occur when species are misidentified so that a species is recorded as detected when a site is unoccupied. Bias in estimators of occupancy, colonization, and extinction can be severe when false positives occur. Accordingly, we propose models that simultaneously account for both types of error. Our approach can be used to improve estimates of occupancy for study designs where a subset of detections is of a type or method for which false positives can be assumed to not occur. We illustrate properties of the estimators with simulations and data for three species of frogs. We show that models that account for possible misidentification have greater support (lower AIC for two species) and can yield substantially different occupancy estimates than those that do not. When the potential for misidentification exists, researchers should consider analytical techniques that can account for this source of error, such as those presented here.

Landscape matrix mediates occupancy dynamics of Neotropical avian insectivores.

In addition to patch-level attributes (i.e., area and isolation), the nature of land cover between habitat patches (the matrix) may drive colonization and extinction dynamics in fragmented landscapes. Despite a long-standing recognition of matrix effects in fragmented systems, an understanding of the relative impacts of different types of land cover on patterns and dynamics of species occurrence remains limited. We employed multi-season occupancy models to determine the relative influence of patch area, patch isolation, within-patch vegetation structure, and landscape matrix on occupancy dynamics of nine Neotropical insectivorous birds in 99 forest patches embedded in four matrix types (agriculture, suburban development, bauxite mining, and forest) in central Jamaica. We found that within-patch vegetation structure and the matrix type between patches were more important than patch area and patch isolation in determining local colonization and local extinction probabilities, and that the effects of patch area, isolation, and vegetation structure on occupancy dynamics tended to be matrix and species dependent. Across the avian community, the landscape matrix influenced local extinction more than local colonization, indicating that extinction processes, rather than movement, likely drive interspecific differences in occupancy dynamics. These findings lend crucial empirical support to the hypothesis that species occupancy dynamics in fragmented systems may depend greatly upon the landscape context.

Batrachochytrium salamandrivorans (Bsal) not detected in an intensive survey of wild North American amphibians.

The salamander chytrid fungus (Batrachochytrium salamandrivorans [Bsal]) is causing massive mortality of salamanders in Europe. The potential for spread via international trade into North America and the high diversity of salamanders has catalyzed concern about Bsal in the U.S. Surveillance programs for invading pathogens must initially meet challenges that include low rates of occurrence on the landscape, low prevalence at a site, and imperfect detection of the diagnostic tests. We implemented a large-scale survey to determine if Bsal was present in North America designed to target taxa and localities where Bsal was determined highest risk to be present based on species susceptibility and geography. Our analysis included a Bayesian model to estimate the probability of occurrence of Bsal given our prior knowledge of the occurrence and prevalence of the pathogen. We failed to detect Bsal in any of 11,189 samples from 594 sites in 223 counties within 35 U.S. states and one site in Mexico. Our modeling indicates that Bsal is highly unlikely to occur within wild amphibians in the U.S. and suggests that the best proactive response is to continue mitigation efforts against the introduction and establishment of the disease and to develop plans to reduce impacts should Bsal establish.

ESTIMATING OCCURRENCE, PREVALENCE, AND DETECTION OF AMPHIBIAN PATHOGENS: INSIGHTS FROM OCCUPANCY MODELS.

Understanding the distribution of pathogens across landscapes and their prevalence within host populations is a common aim of wildlife managers. Despite the need for unbiased estimates of pathogen occurrence and prevalence for planning effective management interventions, many researchers fail to account for imperfect pathogen detection. Instead raw data are often reported, which may lead to ineffective, or even detrimental, management actions. We illustrate the utility of occupancy models for generating unbiased estimates of disease parameters by 1) providing a written tutorial describing how to fit these models in Program PRESENCE and 2) presenting a case study with the pathogen ranavirus. We analyzed ranavirus detection data from a wildlife refuge (Maryland, US) using occupancy modeling, which yields unbiased estimates of pathogen occurrence and prevalence. We found ranavirus prevalence was underestimated by up to 30% if imperfect pathogen detection was ignored. The unbiased estimate of ranavirus prevalence in larval wood frog (Lithobates sylvaticus; 0.73) populations was higher than in larval spotted salamander (Ambystoma maculatum; 0.56) populations. In addition, the odds of detecting ranavirus in tail samples were 6.7 times higher than detecting ranavirus in liver samples. Therefore, tail samples presented a nonlethal sampling method for ranavirus that may be able to detect early (nonsystemic) infections.

Evidence that climate sets the lower elevation range limit in a high-elevation endemic salamander.

A frequent assumption in ecology is that biotic interactions are more important than abiotic factors in determining lower elevational range limits (i.e., the "warm edge" of a species distribution). However, for species with narrow environmental tolerances, theory suggests the presence of a strong environmental gradient can lead to persistence, even in the presence of competition. The relative importance of biotic and abiotic factors is rarely considered together, although understanding when one exerts a dominant influence on controlling range limits may be crucial to predicting extinction risk under future climate conditions. We sampled multiple transects spanning the elevational range limit of Plethodon shenandoah and site and climate covariates were recorded. A two-species conditional occupancy model, accommodating heterogeneity in detection probability, was used to relate variation in occupancy with environmental and habitat conditions. Regional climate data were combined with datalogger observations to estimate the cloud base heights and to project future climate change impacts on cloud elevations across the survey area. By simultaneously accounting for species' interactions and habitat variables, we find that elevation, not competition, is strongly correlated with the lower elevation range boundary, which had been presumed to be restricted mainly as a result of competitive interactions with a congener. Because the lower elevational range limit is sensitive to climate variables, projected climate change across its high-elevation habitats will directly affect the species' distribution. Testing assumptions of factors that set species range limits should use models which accommodate detection biases.

Quantifying climate sensitivity and climate-driven change in North American amphibian communities.

Changing climate will impact species' ranges only when environmental variability directly impacts the demography of local populations. However, measurement of demographic responses to climate change has largely been limited to single species and locations. Here we show that amphibian communities are responsive to climatic variability, using >500,000 time-series observations for 81 species across 86 North American study areas. The effect of climate on local colonization and persistence probabilities varies among eco-regions and depends on local climate, species life-histories, and taxonomic classification. We found that local species richness is most sensitive to changes in water availability during breeding and changes in winter conditions. Based on the relationships we measure, recent changes in climate cannot explain why local species richness of North American amphibians has rapidly declined. However, changing climate does explain why some populations are declining faster than others. Our results provide important insights into how amphibians respond to climate and a general framework for measuring climate impacts on species richness.