You are here:
On the importance of incorporating sampling weights in occupancy model estimation
Citation:
Irvine, K., T. Rodhouse, W. Wright, AND Anthony R. Olsen. On the importance of incorporating sampling weights in occupancy model estimation. The Wildlife Society Meeting, Raleigh, NC, October 15 - 19, 2016.
Impact/Purpose:
Estimating occupancy rates for wildlife species based on data from a survey design is an important step in wildlife monitoring programs. The simulation study relies on an occupancy model developed for complex survey designs. The nascent North American Bat Monitoring Program is designing such a monitoring program for studying the white nose syndrome occurring in bats. The simulation study reported in this study will help monitoring program managers determine an appropriate monitoring design and occupancy modeling approach.
Description:
Occupancy models are used extensively to assess wildlife-habitat associations and to predict species distributions across large geographic regions. Occupancy models were developed as a tool to properly account for imperfect detection of a species. Current guidelines on survey design requirements for occupancy models focus on the number of sample units and the pattern of revisits to a sample unit within a season. We focus on the sampling design or how the sample units are selected in geographic space (e.g., stratified, simple random, unequal probability, etc). In a probability design, each sample unit has a sample weight which quantifies the number of sample units it represents in the finite (oftentimes areal) sampling frame. We demonstrate the importance of including sampling weights in occupancy model estimation when the design is not a simple random sample or equal probability design. We assume a finite areal sampling frame as proposed for a national bat monitoring program. We compare several unequal and equal probability designs and varying sampling intensity within a simulation study. We found the traditional single season occupancy model produced biased estimates of occupancy and lower confidence interval coverage rates compared to occupancy models that accounted for the sampling design. We also discuss how our findings inform the analyses proposed for the nascent North American Bat Monitoring Program and other collaborative synthesis efforts that propose harnessing disparate data sources into a single modeling framework.