Biometrics & Population Modeling II

Contributed Paper
ROOM: HCCC, Room 22

8:10AM Population Level Inferences Improve with Integration of Opportunistic Presence-Absence Data and Systematic Capture-Recapture Data
Catherine C. Sun; Angela K. Fuller; J. A. Royle
Knowledge about population structure and patterns such as distribution and space-use of wildlife populations over large landscapes are crucial for informing conservation and management. However, across wide spatio-temporal extents, limited resources make it difficult to collect systematic data on individuals at high spatial resolution (i.e, spatial capture-recapture data, SCR), which are ideal for precisely and accurately estimating population-level patterns. Furthermore, while citizen science offers an approach to collecting large, opportunistic presence-absence (P-A) datasets on species across wide spatio-temporal extents, analytical methods that integrate opportunistic and systematic data to estimate population patterns are lacking. We developed a spatially explicit integrated population model (IPM) that unites systematic SCR data with opportunistic species P-A data collected at larger spatio-temporal extents to estimate parameters including population density and demographic rates such as survival and recruitment. We conducted simulations with the IPM, demonstrating how opportunistic data can improve population estimates, yield cost-efficiencies in sampling, and extend the spatio-temporal extent of data collection. We used ecological parameters informed by North American black bear (Ursus americanus) populations, and considered a range of systematic and opportunistic detection probabilities (p=0.05 – 0.60), number of opportunistic locations (25, 50, 100), and number of primary sampling occasions (1 – 4). Across SCR detection probabilities, the precision and accuracy of abundance estimates increased with increasing number of opportunistic locations and detection probabilities, decreasing root mean squared error (RMSE) by as much as 40%. Notably, the addition of opportunistic P-A data improved abundance estimates when systematic SCR data were temporally sparse. We discuss the potential for such an integrated data collection and analysis to inform wildlife conservation and management, using a citizen science project developed in New York to collect P-A data on black bears as an example.
8:30AM Use of Informative Priors in Bayesian Occupancy Models to Improve Understanding of Habitat Selection in Declining Species
Laura E. D’Acunto; Patrick A. Zollner; Karly A. Rushmore; Benjamin P. Pauli
Bayesian approaches in ecological modeling have increased in popularity, but use of informative priors in modeling is still relatively rare. Despite the conceptual advantage of using informative priors, there is a lack of guidance or examples of the varied uses informative priors can have for ecological models. We used data from established hierarchical single-species occupancy models for 3 species of bat to parameterize informative priors for habitat selection models for a new dataset in a separate landscape after precipitous population declines. We used resulting model precision and accuracy to test three hypotheses developed under the framework of the ideal free distribution theory: 1) species habitat selection would remain consistent despite population declines, 2) species habitat selection would filter to the most optimal sites as a result of population declines, 3) species habitat selection would change randomly as a result of population declines. We found that for two of the three species, the best models were those using informative priors assuming a change in habitat selection based on population decline. This provides evidence that bat species may be filtering only to optimal habitat as their densities decline. This result provided additional evidence of interspecific spatial foraging competition in North American bat species. We showed that informative priors could be used successfully in a hypothesis testing framework to elicit important ecological insights for imperiled species.
8:50AM Untangling Spatial Patterns in Occupancy Models: Residuals to Separately Assess Detection and Presence
Wilson J. Wright; Kathryn M. Irvine; Megan D. Higgs
Occupancy models are widely applied to estimate species distributions, but there are few methods to assess and evaluate these models. Appropriate model assessments uncover inadequacies and, if done properly, allow for deeper ecological insight by exploring structure not accounted for by a model. Spatial structure is one aspect that requires consideration because it is likely ubiquitous in occupancy studies and failing to adequately model spatial correlation can produce misleading inferences. Residual diagnostics using Moran’s I allow for identifying unaccounted for spatial correlation after fitting a model. We introduce new occupancy model residual definitions that, unlike alternative options, do not condition on naive occupancy. Further, with separate residuals for occupancy and detection, sequential model assessments can distinguish correlation in the detection process from that in the occupancy process. In simulations, we found our proposed residuals outperformed alternative definitions in identifying spatial correlation among detection probabilities. The Moran’s I assessments for example analyses of silver-haired (Lasionycteris noctivagans) and little brown (Myotis lucifugus) bat datasets provided no evidence of residual spatial correlation among occupancy probabilities after accounting for the covariates included in an initial model. Using the residuals we defined for the detection process, however, did suggest residual spatial correlation for this model component. After expanding the initial model to include spatial correlation among detection probabilities, subsequent Moran’s I assessments did not indicate residual spatial correlation. Most applications of spatial occupancy models focus on correlation among occupancy probabilities, but correlation among detection probabilities is also possible and can bias occupancy coefficient estimates. While our examples focused on exploring unexplained spatial correlation, residuals can be used to assess other aspects of model fit as well. Targeting specific model assumptions using carefully chosen residual diagnostics is valuable for any analysis, and we remove previous barriers for occupancy analyses — lack of examples and practical advice.
9:10AM Wide-Ranging Individuals Lead to False Positive Detections and Upwardly-Biased Occupancy Estimates
Gavin M. Jones; William J. Berigan; Sheila A. Whitmore; R. J. Gutiérrez; M. Z. Peery
Occupancy models are fundamental for assessing population status and responses to environmental change; models accounting for false positive detections are increasingly recognized as key to avoiding upward biases in occupancy estimates. While the sampling process (e.g., species mis-classification) is often considered their primary source, false positive detections may also arise from the inherent movement ecology of organisms such as wide-ranging individuals that are detected in multiple, otherwise unoccupied sampling units. To quantify false positive detections rates and occupancy bias resulting from wide-ranging movements, we used GPS and detection/non-detection data of color-marked individuals in a model territorial species – the spotted owl (Strix occidentalis) – over 13 years on a large study area. Thirty-one of 36 GPS-marked owls (86%) used multiple nest/roost areas (the sampling unit for detection/non-detection surveys), and 11 owls (30%) used five or more. In occupancy analyses, 8% of all detections were confirmed false positives of color-marked wide-ranging individuals and 20% of all detections were potential false positive detections. On average, failing to account for false positive detection probability (fp=0.11, 95% C.I.=0.096-0.126) upwardly biased occupancy estimates by a factor of 1.29 (95% C.I.=1.23-1.34). However, in the year following a large, severe fire that affected 30 of 84 owl territories on our study area, failing to account for false positives upwardly biased occupancy by a factor of 1.65 (95% C.I.=1.17-2.12). When not explicitly accounted for, wide-ranging territorial individuals can generate upward biases in occupancy that are further exacerbated following large disturbances when surviving individuals exhibit exploratory movement behaviors in newly altered landscapes. Researchers should minimize the potential for false positive detections through study design, quantify potential false positive detections using field data when possible, and explicitly model false positive errors or risk upwardly biased occupancy estimates and erroneous inferences about population dynamics and conservation status of sensitive species.
9:30AM Efficacy of Evening Point Counts for Estimating Occupancy
Paige F. B. Ferguson; Neil A. Gilbert
The point count has been a methodological mainstay in ornithological research, with researchers typically conducting point counts in the 3–4 hours following sunrise, the interval of maximum avian activity. Avian activity peaks again in the evening, but few studies have examined the efficacy of evening surveys for ecological models. Our objective was to compare occupancy and detection probability estimates from morning point counts with estimates from evening point counts. Knowing whether evening surveys could complement morning surveys is important for designing studies when researchers may want to maximize data collection during a short time period. We conducted morning and evening point counts at 46 sites in the Black Belt Ecoregion of Alabama. Sites were along secondary roadsides, at least 1km apart, and selected using stratified random sampling. We performed surveys from 23 May—27 June 2017 and conducted morning and evening surveys at a site within 48 hours. Using the morning survey data, we fit 5 candidate models with land cover covariates for each of 20 avian species. We used Bayesian occupancy models that account for false positive and false negative detections. We used BIC for model selection and then fit the top model using the evening survey data. To compare estimates from morning surveys against estimates from evening surveys, we measured the overlap of occupancy probability posterior distributions and detection probability posterior distributions. Our results indicated that, for most species, evening detection probabilities were lower and less precise than morning detection probabilities. This contributed to less precise evening occupancy probabilities. For some species, differences in morning and evening occupancy probability posterior distributions contributed to different posterior means. In our study, evening point counts appeared most appropriate for Loggerhead Shrike (Lanius ludovicianus), Indigo Bunting (Passerina cyanea), and Orchard Oriole (Icterus spurius). We discuss recommendations for using evening point counts.


Contributed Paper
Location: Huntington Convention Center of Cleveland Date: October 9, 2018 Time: 8:10 am - 9:50 am