Avian Point Count Surveys: Estimating Components of the Detection Process

Abstract

Point count surveys of birds are commonly used to provide indices of abundance or, in some cases, estimates of true abundance. The most common use of point counts is to provide an index of population abundance or relative abundance. To make spatial or temporal comparisons valid using this type of count requires the very restrictive assumption of equal detection probability for the comparisons being made.

We developed a multiple-independent observer approach to estimating abundance for point count surveys as a modification of the primary-secondary observer approach. This approach uses standard capture-recapture models, including models of inherent individual heterogeneity in detection probabilities and models using individual covariates to account for observable heterogeneity in detection probabilities. Two-observer models provided negatively biased estimates because they do not account for individual heterogeneity in detection probabilities. Models accounting for individual heterogeneity are always selected as the most parsimonious models for this data type.

We also developed a time of detection approach for estimating avian abundance when birds are detected aurally, which is a modification of the time of removal approach. This approach requires collecting detection histories of individual birds in consecutive time intervals and modeling the detection process using a capture-recapture framework. This approach incorporates both the probability a bird is available for detection and the probability of detection given availability. Analyses presented demonstrate the importance of models accounting for individual heterogeneity in detection probabilities. We recommend time of detection point count surveys be designed with four or more equal intervals.

We also present a multiple species modeling strategy since many point count surveys collect data on multiple species and present the approach for distance sampling, multiple observer, and time of detection approaches. The purpose of using a multiple species modeling approach is to obtain more parsimonious models by exploiting similarities in the detection process among species. We present a method for defining species groups which leads to an a priori set of species groups and associated candidate models. Multiple species models worked well and in many cases gave more parsimonious models than a species specific modeling approach, especially for the multiple-observer and time of detection approaches. Parameter estimates for multiple species models are more precise than single species models. We recommend this approach for all situations where data on multiple species is collected.

Finally, we present a method for estimating the availability probability of birds during a point count based on singing rate or detailed singing time data. This approach requires data collected in conjunction with point count surveys that describe the singing rates or singing time distribution of the bird population of interest. The singing rate approach requires the assumption that an individual bird sings following a random process but rates may vary between birds. We modeled this using a finite-mixture Poisson model. The singing time approach is a nonparametric approach and does not require this restrictive assumption. Analysis of Ovenbird singing rate data demonstrates the importance of accounting for availability bias when estimating abundance, especially as count lengths get short. We recommend this approach when 'snapshot' type counts are necessary.

Analyses presented throughout this thesis demonstrate the importance of accurately modeling the detection process to estimate abundance. The importance of accounting for individual heterogeneity in detection probabilities was evident in every chapter. Using a point count method that accounts for individual heterogeneity is crucial to estimating abundance effectively and making valid spatial, temporal and species comparisons.