Spatial Modeling of Detection and Abundance from Count Surveys of Animal Populations

Bayesian models;detection probability;bird surveys;CAR models;Mexican hat density;distance sampling

Webster, Raymond Anthony ; Kenneth H. Pollock, Committee Chair,Sujit K. Ghosh, Committee Member,Cavell Brownie, Committee Member,Kevin Gross, Committee Member,Webster, Raymond Anthony ; Kenneth H. Pollock ; Committee Chair ; Sujit K. Ghosh ; Committee Member ; Cavell Brownie ; Committee Member ; Kevin Gross ; Committee Member

University:North Carolina State University

关键词: Bayesian models;    detection probability;    bird surveys;    CAR models;    Mexican hat density;    distance sampling;   

Others  :  https://repository.lib.ncsu.edu/bitstream/handle/1840.16/3133/etd.pdf?sequence=1&isAllowed=y

美国|英语

来源: null

【 摘 要 】

When analyzing data from surveys of animal populations, it has been common in the past to ignore important factors such as variation in animal detection probabilities across space, and spatial dependence in animal density.We present a unified framework for modeling animal survey data collected at spatially replicated survey sites in the form of repeated counts, "removal" counts, or "capture" history counts, that simultaneously models spatial variation in density and variation in detection probabilities due to changes in covariates across the landscape.The models have a complex hierarchical structure that makes them suited to Bayesian analysis using Markov chain Monte Carlo (MCMC) algorithms.To ensure that these algorithms are computationally efficient, we use conditional autogressive (CAR) models for modeling spatial dependence.We apply our models to two examples of animal survey data.In the first, an intensive repeated count survey of juvenile Coho Salmon in McGarvey Creek, Northern California, we detected moderate spatial dependence in density, and models which account for spatial dependence produced more precise predictions at unsurveyed habitat units, and thus more precise estimates of total stream abundance, than models which assumed spatial independence.Through a small simulation study, we show that ignoring heterogeneity in detection probabilities can lead to significant underestimation of total abundance.However, inclusion of heterogeneity using a random effect in the detection component of the model can lead to problems in Bayesian MCMC modeling for typical survey designs, and for this reason we stress the importance of accounting for heterogeneity by incorporating covariates in modeling detection probability.In our second example, we consider a large survey of birds in the Great Smoky Mountains National Park.We fit models to the three types of survey data, repeated counts, "removal" counts, and "capture" history counts.Our methods lead to maps of predicted relative density which are an improvement over those that would follow from ignoring spatial dependence.Modeling shows that variation in detection probability can also affect inference, particularly when a species is relatively difficult to detect.Our work also highlights the importance of good survey design for bird species modeling.We point out that these types of bird survey data, particularly removal and capture-recapture counts (which require individual birds to be identified), are prone to errors in bird identification.Although we obtain similar results for all three types of survey data, which implies that the effect of identification errors may be small, the consequences of such errors in the data requires further investigation. Finally, we present parametric models for combined distance and capture-recapture survey data from both line and point transect surveys that allow for two types of animal movement: permanent avoidance or attraction to a transect, or temporary displacement of animals in the vicinity of a transect.The models have a simple form, with parameters that quantify the impact transects and observers have on local density.We combine these density models with logistic-linear models for detection probability using the likelihood framework of Borchers et al. (1998) for combined distance and capture-recapture data.This allows us to separately estimate the parameters of both the density and detection components of the model, which is not possible using the standard methods of distance sampling.Through a simulation study, we show that, provided sufficient animals are detected, the model parameters have little bias, and lead to improved estimates of density over a simple uniform density model, particularly for line transect surveys.Model selection by AIC generally chooses the correct density model.We apply our models to the Great Smoky Mountains bird survey data, and find some evidence of observer effects on local bird density.