If you have any problems related to the accessibility of any content (or if you want to request that a specific publication be accessible), please contact firstname.lastname@example.org.
Applications and Extensions of Occupancy Estimation and Modeling Using Ecological Monitoring Data
AdvisorAlbright, Thomas P
AltmetricsView Usage Statistics
Ecological monitoring data are critical in estimating populations, distributions, and other parameters. These data are collected by a variety of agencies, non-governmental organizations, volunteer networks, and private individuals as part of coordinated programs operating with a variety of conservation-related goals. The programs face numerous challenges, often leading to shortcomings in data quality and utility. Likewise, the species targeted by these programs may be imperfectly detected due to accessibility, rarity, cryptic behaviors, or other issues resulting in negative surveys (non-detections) despite the presence of a species.A suite of approaches and statistical tools have been developed in the last two decades employing data from repeated visits or other sampling designs allowing the estimation of both observation (e.g., detection) and state processes (e.g., presence or absence). These approaches have improved the ability of researchers, monitoring agencies, and others to more reliably (1) estimate species distributions, abundances, or diversity measures, (2) estimate changes in these variables over time or space, (3) elucidate relationships with environmental covariates, and (4) generate predictions for imperfectly-sampled locations, non-sampled locations, or projected future conditions. However, even when accounting for observation processes, other sources of variation may result in inaccurate estimation or inference about relationships underlying detection, presence or abundance, and occupancy or population dynamics through time and space. In my first analysis chapter, I address one such source—within- and between-group effects—resulting from variation in predictor covariates across levels such as individuals or locations. These effects may be modeled by first decomposing covariates (using simple arithmetic) into their within-group means (between-group component) and within-group deviation from these means (within-group component). As a case study, I employ a multi-species, multi-year data set of detection vs. non-detection records (i.e., capture vs. non-capture) from the well-known Biological Dynamics of Forest Fragments Project in the Brazilian Amazon. I decomposed capture effort (in mist-net hours) into within- and between-site, -year, and -site-year components and compared support for, and effects within, multi-season occupancy models including these decomposed effort covariates vs. models employing the raw covariate of capture effort. I found at least marginal model improvement (ΔAICC > 0) when employing decomposed effort covariates in over 90% of species and notable model improvement (ΔAICC > 2) in over 50% of species. Species exhibiting improved prediction via decomposed covariates had effort-associated effect sizes in their best model that were on average 22% greater than raw covariates. These differences were most pronounced in between-site and within-year effects of effort and suggested detectability related to site characteristics (e.g., area or isolation) or year-to-year population variation were better addressed via decomposed covariates.As a practical application of occupancy estimation and modeling to a species of conservation concern, I considered two research areas employing long-term monitoring data of the Lesser Prairie-chicken (Tympanuchus pallidicinctus; LPCH) in modeling behavior and distributions through time and space. The first research area focused on heterogeneity in lek attendance of LPCH (the presence of males on communal breeding grounds given their presence in an area) using information associated with sampling occasion (lek survey) and landscape attributes. This area of interest is important given the widespread use of lek surveys as a monitoring and management tool for the LPCH and other prairie grouse. Using generalized linear models of lek-year combinations with at least two surveys and one observation of lek attendance, I found male LPCH lek attendance varied in relation to lek size, survey conditions (survey date, daily wind speeds, and estimated duration), and characteristics of the landscape (herbicide-treated area) and the previous winter’s weather (temperature and precipitation). Lek attendance estimates suggested survey coordinators and field crews should consider the use of multiple surveys, survey-related conditions, and environmental characteristics potentially influencing the probability of observing males on leks to improve estimates of LPCH populations or distributions.The second area of focus in which I employed LPCH lek survey data was in modeling the dynamics of site occupancy through time and space. The importance of dynamic species distribution models (SDMs) as monitoring and management tools has grown in the past decade with the availability of statistical tools and environmental data (e.g., gridded weather or remotely-sensed landscape data sets). However, considerations of (1) a species’ non-equilibrium with its environment, (2) imperfectly-collected monitoring data, and (3) imperfect species detection are less commonly addressed in SDMs. Using multi-season occupancy models accounting for variation in lek attendance, I constructed dynamic SDMs employing three covariate groups associated with demographic rates (e.g., reproductive success) and expected to be predictive of occupancy dynamics. These groups were population indices (neighborhood occupancy and lek size), landscape characteristics (vegetation, herbicide treatment, and roads within 3.2 km of sites), and weather (seasonal temperature anomalies, precipitation anomalies, and climate indices). The most-supported model included covariates from each of the three groups, provided good classification accuracy, and produced predictions through time illustrating complex synergies between internal population processes (e.g., dispersal), landscape changes, and inter-annual weather variation.Equipped with new statistical tools, monitoring data, environmental data sets, and the improved inference possible when they are carefully integrated, ecological and biogeographical researchers are ideally situated to identify novel patterns, elucidate poorly-understood processes, and offer recommendations to managers and other decision-makers. My approaches and conclusions build off a growing literature focused on improving inference using modern statistical approaches employing ecological monitoring data. In this dissertation, I provide novel applications and extensions in occupancy estimation and modeling that can improve inference across a variety of species and environmental contexts. Future research applying occupancy models or emerging spatiotemporal statistical approaches employing robust samples from similar or markedly different study systems will no doubt improve our ability to estimate and understand drivers of species occurrence or abundance through space and time.