Overview
I am a statistical ecologist interested in the development and application of novel statistical methods for analyzing ecological data and I enjoy working with state and federal natural resource agencies to use these methods to answer applied research questions. Below is an overview of my work’s major themes and corresponding projects.
Demography
Contributions of Weather Patterns and Land Cover Changes to Waterfowl Demography in the Prairie Pothole Region
Separating the effects of change in anthropogenic land use and climate change is of critical importance for understanding temporal changes in species abundance. However, land-use changes and weather patterns operate at vastly different scales of influence, both in time and in space. Land-use changes and climactic conditions have both been identified as possible contributors to declines in the breeding population of waterfowl in the Prairie Pothole Region (PPR), which have been monitored since 1955 by the FWS Breeding Waterfowl and Habitat Survey. To disentangle the effects of land-use changes and climactic conditions, we are developing a hierarchical process model that separates small-scale, within-year determinants of settling patterns from large-scale demographic drivers of change in abundance. We will also explicitly model the detection process and the process by which annual precipitation and temperature regimes contribute to the number of available wetlands, allowing us to evaluate the direct and indirect effects of climate on waterfowl. In addition, we will be making single species inference on species of concern (e.g., northern pintails) and across species, to determine if responses to land use and climate vary by life history traits.
Harvest Rates and Hunter Selectivity of White-tailed Deer
Accurate estimates of annual survival and harvest rates are important for the management of many game species. These estimates are often obtained by capturing and marking a subset of the population and relying on tag-returns during the hunting season to estimate harvest rate. However, many game species can only be captured in winter, which may be many months before the hunting season. Mortality during these intervening months is a violation of tag-recovery models and will result in underestimates of the harvest rate. We developed a model that incorporates known-fate data (e.g., radio-collars) into a tag-return model to correctly account for tagging-to-hunting season mortality and evaluated combinations of radio-collars and reward tags to determine optimal sample size allocations (Buderman et al. 2014a). Because some states use radio-collared deer to estimate harvest rates, we then used this model to determine whether Pennsylvania hunters were more or less likely to harvest a radio-collared deer (Buderman et al. 2014b).
Optimal Sampling and Abundance Estimation for Salamanders
Salamanders are often used as indicators of forest health, making accurate estimates of population size important for conservation planning. Researches typically survey for terrestrial salamanders by performing diurnal searches for individuals resting under moist rocks and logs or nocturnal searches for individuals foraging on wet ground surfaces. We evaluated differences in the probability of initial capture and recapture of individuals using the two search methods and the effect of search method on abundance estimates. Initial capture probabilities were higher during nocturnal searches, while recapture probabilities were higher during diurnal searches; this typically resulted in slightly higher abundance estimates from nocturnal surveys. We hypothesize that the two search methods sample different subsets of the population. We also found large annual variation in abundance estimates when we included a search method effect, indicating that multi-year monitoring may be necessary for an accurate assessment of the salamander population (Buderman and Liebgold 2012).
Animal Movement
Long-distance Movement of Canada Lynx
From 1999-2006, 218 wild-caught Canada lynx were reintroduced by Colorado Parks and Wildlife to the San Juan Mountains of Colorado. To monitor survival and reproduction, each individual was fitted with a telemetry collar upon release. Based on these demographic parameters, the reintroduction was considered a success. As a by-product of this reintroduction, we now have a lynx location data set that is unprecedented for the United States in both sample size and temporal span. However, the data were collected at a coarse temporal resolution and contain multiple data sources with differing and unusual error structures, prohibiting the use of other contemporary movement models. Therefore, we developed a functional movement model in a Bayesian framework that can account for multiple data sources and provide inference on both the location of an individual and derived behavioral quantities related to the movement path (such as speed, residence time, and tortuosity; Buderman et al. 2016; also featured on methods.blog). We then extended this model to the population-level and summarized quantities of interest, such as the timing and duration of long-distance movement, correlations between movement and landscape characteristics, philopatry, and the effect of reintroduction on movement (Buderman et al. In Review).
Time-varying Behavior of Mountain Lions
The second half of my dissertation concerns quantifying drivers of movement for mountains lions in the Front Range of Colorado. The Front Range is a mountain range that runs north-south, just west of the most populous urban corridor in Colorado. Mountain lions frequently move between the protected natural areas and open spaces into more suburban and urban areas. Since 2007, Colorado Parks and Wildlife has collared mountain lions with GPS collars and obtained locations approximately every 3 hours. This fine-scale temporal data, in conjunction with a model developed by Hanks et al. 2015, will allow us to quantify how the response of mountain lions to covariates, such as kill site location, prey abundance, and housing development, varies temporally (e.g., time of day and time since last kill).
Methodological Development
Much of my work begins with a methodological issue (e.g., optimal sampling designs, utilizing historical data, incorporating multiple data sets) and ends with an application to a particular data set or question of management interest (such as those listed above).
Historical Data Sets
Wildlife data sets are often time consuming and costly to collect and often it may be more cost-effective or logistically reasonable to use previously collected data sets to investigate new ecological questions. However, since the data were collected for a different purpose, we may need to develop new methodologies to account for characteristics of the data that would not exist in an ideal data set (Buderman et al. 2016).
Integrated Data Models
Different kinds of data are often collected on the same population over a time period of interest. These data sets may be collected by a single group or multiple groups studying different aspects of the same system. Integrating multiple data sets can reduce uncertainty and bias when they arise from the same underlying process and can be integrated using either a likelihood (Buderman et al. 2014a) or a Bayesian framework (Buderman et al. 2016).
Optimal Sampling Designs
As mentioned previously, collecting multiple data types can often result in improved parameter estimates. However, there is a trade-off between associated costs of monitoring and the information content of the data collected. Therefore, given some desired level of precision and cost, we can use simulations to determine the optimal sample size for each data type (Buderman et al. 2014a). Monitoring methods can also be evaluated for their ability to detect trends (Buderman and Liebgold 2012) and sample a representative subset of the desired population (Buderman and Liebgold 2012; Buderman et al. 2014b).
Computational Skills
Given my interest in applied statistical ecology, I have experience with many of the software programs used by ecologists and wildlife biologists, such as MARK, DISTANCE, and SURVIV. I have experience using ArcGIS and Microsoft Access; I am extremely knowledgeable regarding data analysis and visualization in R, including using R as a GIS, and teach workshops on the subject. Depending on the application, I use both the R interface for JAGS (rjags) and code my own MCMC samplers for Bayesian analysis.