Kernal Density Methods for Estimation of Nonstandard Detection Functions

• Activity Number: 572
• Type: Topic Contributed
• Date/Time: Wednesday, August 4, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): James Griswold*+ and Trent McDonald and Michelle Borassou-Stahl
• Companies: Western EcoSystems Technology, Inc. and Western EcoSystems Technology, Inc. and Western EcoSystems Technology, Inc.

Abstract

For distance sampling, standard distance analysis methods for estimating probability of detection utilize semi-parametric models to force estimated detection functions to be highest at the minimum available sighting distance and to decrease monotonically out to the maximum sighting distance. These models can be fitted using the program DISTANCE (Thomas et al. 2006). However, lower detection probabilities at the minimum sighting distance has been documented for surveys from fast moving aircraft (e.g., Becker and Quang 2009, Nielson et al. 2010). Closer objects are in view for less time and thus can be more difficult to detect. Accordingly, a more appropriate analysis might utilize a non-monotonic kernel estimator to model the shapes of detection functions. We discuss this approach, and identify its advantages and disadvantages. Simulations indicated that the kernel method performed better than DISTANCE at the low densities we experienced in our application.