Modeling Communities While Accounting for Imperfect Detection

ROOM: HCCC, Room 21
Community models consist of the joint modeling of data from multiple species, allowing estimation of species-specific parameters while sharing information across species under the assumption that parameters come from a common, underlying distribution. This form of information sharing can improve parameter estimates for data poor species in both static (single-season) and dynamic models of species occurrences and abundances. Community models have been combined with different data types (detection/non-detection, repeated counts, distance sampling), allowing researchers to make inference on both species-specific effects as well as aggregated effects of covariates on the entire community while accounting for imperfect detection. Combined with data augmentation, community models facilitate mechanistic estimation of species richness and other diversity metrics. The community modeling framework is increasingly popular in wildlife research as novel methods such as camera trapping, high definition video surveys or acoustic recorders facilitate data collection for entire communities of species simultaneously. Consequently, the field has seen rapid methodological development. Recent extensions include multi-site models to investigate geographic variation in richness; incorporating phylogenetic relationships among species; addressing temporary emigration; allowing for latent group structure in the community; and exploring species interactions. This symposium provides an overview of recent methodological development and innovative applications of the community modeling framework.

12:50PM Introduction to Community Models Accounting for Imperfect Detection
  Elise Zipkin; Rahel Sollmann
Multi-species, or community, models combine data from all available species into a single analytical framework. Recent developments in multi-species modeling have improved the ability to make inferences about species and communities based on species-specific models of occurrence or abundance, integrated within a hierarchical modeling framework. This modeling framework offers advantages to inference about species occupancy, distributions, abundance and/or community composition over typical approaches by accounting for both species-level effects and the aggregated effects of environmental and other covariates on a community as a whole. Integrating data from an entire community improves estimates for all species, including those that are rare or were observed infrequently. Community models have been used for inference on a variety of topics, including estimation of richness and community overlap, construction of species accumulation curves, and in determining the influence of habitat and landscape variation on species richness and composition. The community modeling framework is increasingly popular in wildlife research as novel methods such as camera trapping, high definition video surveys, and acoustic recorders facilitate data collection for entire communities of species simultaneously. Consequently, the field has seen rapid methodological development and a number of novel applications. This talk will provide a review of community modeling and the importance of accounting for imperfect and variable detection in the context of community analyses. We will discuss the significance of this modeling framework for ecological research, highlighting both past works and discussing future potentials.
1:10PM Model Selection for Community Models
  Kristin M. Broms
Multi-species occupancy models (MSOMs) are one way to model communities while accounting for imperfect detection, and the models have many advantages. For example, they produce unbiased occupancy and detection estimates from a minimal number of sites; the inclusion of random effects in the model help inform distributions for species with few detections; the framework can be used to gain site-specific, species-specific, and community-wide insight; and species richness and other metrics of biodiversity can be derived from model output. While the MSOM offers many advantages, its Bayesian framework makes model selection and assessment more difficult for researchers who are accustomed to using AIC and linear regression diagnostics. The selection and assessment of best-fitting and best-predicting Bayesian models are areas of active research and there are multiple options available to researchers. While information criterion formulas have been proposed to compare Bayesian models, cross-validation methods have been proven to be effective in regard to selecting a best-predicting model. In the presentation, we will explore cross-validation methods for MSOMs and discuss the other alternatives.
1:30PM A multi-region model for inference about geographic variation in community size and structure.
  Chris Sutherland; Simone Tenan
An enduring challenge in ecology is to understand what drives spatial variation in the size and structure of communities. The ability to count the number of species present at a location is hindered by the fact that not all species are equally detectable, and invariably some go completely undetected. This makes comparing species richness across distinct spatial units (or regions) problematic as sources of error are usually unaccounted for in simple enumerations of species. We present a hierarchical multi‐region community model that allows for direct modelling of trait‐based patterns of species richness along environmental gradients by partitioning communities into ecologically relevant strata (e.g. guilds). This framework allows for joint estimation, and explicit hypothesis testing, of region‐specific community size and structure. We demonstrate the multi-region community model using camera trapping data collected by the Tropical Ecology Assessment and Monitoring (TEAM) Network. Specifically, we investigate the functional composition and vulnerability of tropical forest mammal communities.
1:50PM Modeling Species Interactions in Multi-Species Occupancy Models.
  Mathias W. Tobler; Marc Kéry; Thomas Sattler; Peter Knaus
Spatiotemporal patterns in biological communities are driven by environmental factors and biotic interactions. They are naturally described by combining models for all species in the community into a single model. Two important considerations in such multi-species occupancy models or joint species distribution models (JSDMs) are presence/absence measurement errors and interactions between species. Virtually all JSDMs have included either one or the other, but never both features simultaneously, even though both measurement errors and species interactions may be essential for achieving unbiased inferences about communities. We developed two JSDMs for modeling pairwise species interactions while accommodating imperfect detection; one with a latent variable and another with a multivariate probit approach and evaluated their performance with three simulation studies. We also illustrate our models with a large Atlas data set on 62 passerine bird species in Switzerland. Under a wide range of conditions, our new latent variable JSDM with imperfect detection and species interactions yielded estimates with little or no bias for occupancy, occupancy covariates and the species correlation matrix. In contrast, we were unable to estimate correlation for datasets with more than about 10 species with the multivariate probit model. A latent variable model that ignores imperfect detection produced correlation estimates with a consistent negative bias. The analysis of the Swiss passerine dataset illustrated well how not accounting for imperfect detection will lead to negative bias in occupancy estimates and to attenuation in the estimated covariate coefficients in a JSDM. In addition, any covariate affecting detection may cause spurious patterns in the estimated species correlation matrix.
2:10PM Multi-Species Hierarchical Modeling Reveals Variable Responses of African Carnivores to Management Alternatives
  Matthew T. Farr; David S. Green; Kay E. Holekamp; Gary J. Roloff; Elise F. Zipkin
Carnivore communities face unprecedented threats from anthropogenic disturbance. Management regimes have variable effects on carnivores where species may persist or decline in response to changes to the ecosystem. Using a hierarchical multi-species modeling approach, we examined the effects of alternative management regimes (i.e., active vs. passive enforcement of regulations) on carnivore densities at both species and community levels in the Masai Mara National Reserve, Kenya. Alternative management regimes have created a dichotomy in ecosystem conditions within the Reserve where active management enforcement maintains low levels of human disturbance in the Mara Triangle and passive management enforcement in the Talek region permits multiple forms of human disturbance. The alternative regimes have variable effects on 11 observed carnivore species. Some species, such as African lions and bat-eared foxes, had higher population density in the Mara Triangle, where regulations are actively enforced. Yet, other species, including black-backed jackals and spotted hyenas, had higher population density in the Talek region where enforcement is passive. Our results indicate that active and passive enforcement of regulations on human activities in protected areas has direct effects on carnivore communities. Multiple underlying mechanisms, including behavioral plasticity and competitive release, are likely causing higher black-backed jackals and spotted hyena population densities in the disturbed Talek region. Yet, high levels of human disturbance negatively affect the majority of carnivore communities producing consequences that likely permeate throughout the rest of the ecosystem. Our multi-species modeling framework reveals that carnivore species do not react to management regimes uniformly, shaping carnivore communities by differentially producing winning and losing species. Some carnivore species may require active management enforcement for effective conservation while others may adapt to passive management enforcement and anthropogenic disturbance, even thriving in such environments. Community approaches to monitoring carnivores should be adopted as single species management may overlook important intra-community variability.
2:30PM Refreshment Break
3:20PM Multiple taxa to multiple regions: biodiversity monitoring in California
  Lindsey Rich; Brett Furnas; Justin Brashares; Stacy Anderson; Steven Beissinger
Maintaining biodiversity in the face of land use and climate change is a paramount challenge, particularly when distributions of many species remain incompletely known. Emerging technologies help address this data deficiency by facilitating the collection of spatially explicit data for multiple species from multiple taxa. California’s Department of Fish and Wildlife (CDFW) initiated multi-taxa, multi-region surveys to capitalize on these newly available tools and improve our understanding of wildlife populations in the state. As part of this effort in 2016, CDFW deployed camera traps and acoustic detectors at over 500 sites across six ecoregions in California (i.e., 200,000km2), ranging from the Mojave Desert to the Klamath Mountains. We identified detections to the species-level and used multispecies occupancy models to estimate the distributions of mammal and songbird species at regional to statewide-scales, and to evaluate community and species-specific responses to climate and habitat variables. We detected 29 mammal (>0.5 kg) and 137 songbird species, with occupancy estimates ranging from 0.01-0.74. Among the species with the highest estimated occupancies were black-tailed jackrabbits and black-throated sparrows in the southern part of California, and western tanagers and black bears in the north. At the statewide level, our model revealed species richness was generally greatest close to forest cover and at mid-elevation gradients. Variation among species’ responses was explained in part by regional differences. The negative influence of temperature, for example, was strongest in the Mojave Desert. Our research demonstrates the utility of using data from visual and acoustic sensors, in combination with multispecies occupancy models, to estimate and evaluate mammal and songbird distributions across multiple spatial scales. We encourage long-term collection of this type of multi-taxa, multi-region data as it would allow managers to detect biodiversity changes, to identify ecological stressors, and to develop, track, and adapt regional and statewide management actions.
3:40PM Successes and Failures in Using Community Models to Inform Management of Public Lands
  Angela White
Occupancy models frequently use detection data on a species to make inferences about habitat suitability as a function of current and potential future environmental conditions. These models are increasingly being used to predict and monitor trends in wildlife populations across continental scales but are too coarse to provide land management agencies with information that can effectively inform management. Here we present the results of several studies on public lands in which multispecies occurrence models were used to develop decision support tools for managers and stakeholders to predict changes in community composition following an environmental disturbance. These models allow managers to attribute specific management activities with biodiversity outcomes and if adopted regionally would provide a framework for setting biodiversity goals across large landscapes and multiple jurisdictional boundaries. We suggest that predictive tools such as these may be among the primary products and benefits of broad-scale monitoring, including the potential to interpret consequences for overall ecosystem function and resilience. However, the efficacy of these models in predicting population changes over space and time is dependent on consistency in species-specific responses to environmental covariates.
4:00PM A New Method for Estimating Multispecies Abundance Using Dependent Double-Observer Surveys
  Jessie Golding; J. Joshua Nowak; Kaitlyn Strickfaden; Victoria J. Dreitz
Multispecies surveys provide a cost-effective way to estimate the abundance of individual members of a biological community. Yet, accounting for sources of error in multispecies surveys can be challenging. Recent advances in estimating abundance of multiple species, such as Bayesian multispecies N-mixture models, account for multiple sources of variation including detection error. However, false-positive errors (misidentification), which are prevalent in multispecies surveys, remain largely unaddressed. The dependent-double observer (DDO) method accounts for detection error and is known to reduce false positives compared to single observer methods because it relies on two observers collaboratively identifying individuals. To date, the DDO method has not been combined with advantages of multispecies N-mixture models. We create a new multispecies dependent double observer abundance model (MDAM), which is a multispecies N-mixture model that incorporates the DDO method. The MDAM uses a hierarchical structure to account for biological and observational processes in a statistically consistent framework while using accurate observations from the DDO method. Simulated and field data analyses show the MDAM accurately estimates abundance of multiple species and can be used to assess differences in abundances of multiple species. Simulations showed the model provides precise and accurate abundance estimates, with average credible interval coverage across 100 simulations of 94.5% for abundance estimates and 92.5% for detection estimates. In addition, 92.2% of abundance estimates had a mean absolute percent error between 0% and 20%, with an overall mean of 7.7%. We also applied the model to field data to estimate the effect of management practices on the abundance of multiple bird species. We suggest that the MDAM and its ability to use the DDO method may be applied in other multispecies surveys, such as winter forest carnivore surveys and other situations where two people are required for surveys due to access or safety concerns.
4:20PM Using multispecies occupancy models to understand the interplay of patch size and life history traits in shaping Southern California small vertebrate communities
  Staci M. Amburgey; David A. Miller; Katy Delaney; Seth Riley; Carlton J. Rochester; Robert N. Fisher
Urbanization impacts species communities as it leads to increased habitat fragmentation and can disrupt historic dispersal corridors, creating patches of potentially suitable habitat with limited movement in between. Community-wide sampling is needed to understand the way species respond to these altered habitats; however, approaches that account for imperfect species detection are needed to prevent underestimation of species presence. We collected data from pitfall traps (n = 750 arrays, t = 3-17 years) across Southern California, sampling 50 species of reptiles, amphibians, and small mammals. We used a hierarchical, multispecies occupancy model to estimate site- and patch-level species richness as a function of overall patch size. Using this model, we obtained species-specific occupancy and detection probabilities at these two spatial scales and calculated species richness at sites and patches. At both the site and patch, species richness was lower in smaller patches with no density compensation (i.e., no filling of empty niches) when species went locally extinct at a site within a patch. We found species-specific variation in the direction of the effect of patch size on occupancy probability. To better understand these species-specific differences, we included life history characteristics (e.g., fecundity, body size) in occupancy models to investigate traits that can make species resilient or sensitive to fragmentation and lead to observed distributional patterns across the gradient of development in Southern California. Community-level analyses that account for imperfect detection in this system will help elucidate the role of urbanization in altering species communities and inform potential mitigation policies in land conservation practices.
4:40PM Using Community Occupancy Models to Quantify Community Diversity
  Jesse F. Abrams; Rahel Sollmann; Andreas Wilting
Estimates of species richness and diversity are central to community ecology and conservation planning. Diversity, however, is a generic term used for a variety of measures that try to incorporate the complex multidimensional properties of a community. It is not uncommon to find that diversity has increased according to one index, but decreased according to another. Measurements of diversity require information on species abundance. For many species, abundance estimation is logistically unfeasible, but using detection/non-detection and occupancy models allows estimation of occurrence. Here, we use simulated data of animal communities to explore the accuracy of occupancy-based diversity measurements. We simulated communities of different composition based on species abundances. We then generated detection/non-detection data for these species and analyzed these with a community occupancy model. We used predictions of species occupancy as a proxy for abundance to calculate occupancy-based diversity indices. We found that occupancy estimates were positively biased in species with low detection probability, resulting in an overestimation of diversity. Applying an occupancy threshold to determine presence led to unbiased estimates of diversity, but different methods to determine the threshold performed better for different diversity indices. To overcome this complexity, we also used diversity profiles, which simultaneously capture the common diversity indices. Although the application of one common threshold to diversity profiles cannot produce an exact match between abundance and occupancy-based profiles, these profiles allowed us to identify differences between our simulated communities. Despite certain limitations our results show that we can glean similar diversity information from occurrence data as from abundance data and that occupancy-based diversity profiles might be a useful tool to compare multiple communities or to monitor diversity trends over time.

Organizers: Rahel Sollmann, University of California Davis, Davis, CA; Elise Zipkin, Michigan State University, East Lansing, MI
Supported by: Biometrics Working Group

Location: Huntington Convention Center of Cleveland Date: October 8, 2018 Time: 12:50 pm - 5:00 pm