OOS 4-6
Integrating count and occupancy data to model population dynamics

Monday, August 10, 2015: 3:20 PM
316, Baltimore Convention Center
Elise Zipkin, Department of Integrative Biology, Michigan State University, East Lansing, MI

Large scale spatial and temporal data are required to estimate species distributions, viabilities, and demographic rates.  As ecological questions become increasingly complex, it is necessary to develop integrated models that make use of data from multiple sources as no one dataset is likely to characterize a species across the complete spatial range of interest.  This is particularly true when interest lies in understanding how demographic rates change relative to climate and/or landscape covariates.  Traditional “integrated models” have focused on combining capture-recapture data with time series count data.  Capture-recapture data are generally expensive and time-intensive to collect.  As such, additional approaches that incorporate “cheaper” data are valuable.  We present an integrated modeling approach to combine time series of count and occupancy data.  The underlying state-process, abundance, is modeled according to an open population n-mixture model (e.g., the Dail-Madsen approach) where population abundance changes over time according to survival and gains (immigration and recruitment).  Count data are used to estimate these parameters.  We then augment the dataset with detection/nondetection data, which can typically be collected more easily at broad spatial scales. 


The resulting integrated population model makes use of both count and detection/nondection data to more precisely estimate recruitment, survivorship, colonization, and extinction rates.  Habitat covariates can be added into the model in cases where data are available over broad spatial extents, allowing for a greater understanding of how demographic parameters and population abundance change over a species’ geographic range.  We demonstrate the data requirements for this modeling approach through a series of simulations across a spectrum of realistic parameter values.  We then illustrate the utility of the approach with an example using wood frog data.