Problems with Avian Point Counts for Estimating Density of Northern Bobwhite—A Case Study

Avian conservation programs such as the North American Landbird Conservation Plan (Rich et al. 2004) and Joint Ventures (U.S. Fish and Wildlife Service, 2010) set population goals for priority bird species such as northern bobwhite Colinus virginianus. These population goals often take the form of population densities. Point counts are widely applied to estimate population densities of many bird species (Wunderle 1994; Hamel et al. 1996). Point counts, as a sampling method, use randomly placed, independent point locations from which to record birds detected by sight or sound. Data collected from point counts may be used to provide population indices or population estimates.

Historically, managers use indices such as morning whistle counts or North American Breeding Bird Survey (U.S. Geological Survey, Patuxent Wildlife Research Center and Environment Canada, Canadian Wildlife Service, 2001) routes to develop trends in northern bobwhite populations (Rosene 1957; Robbins et al. 1989). These indices alone are not robust and cannot be used as surrogates for density estimates (Norton et al. 1961; Applegate 2000; Anderson 2001). Thus, managers are looking to distance-based point counts to provide reliable measures of population density for northern bobwhite and other birds as part of a coordinated monitoring system (Somershoe et al. 2006; Lambert et al. 2009). Point counts for avian species may be used in conjunction with distance sampling theory (Buckland et al. 2001) to estimate density and abundance. Distance sampling is based on the concept that the probability to detect a bird decreases as the distance from the point being surveyed increases, providing a framework for fitting empirical data that represent this inverse relationship to known distributions.

This study uses a large, single-observer data set of point counts from Tennessee USA to examine efficacy of deriving northern bobwhite densities using distance sampling. This approach for northern bobwhite density estimation was piloted in a shorter 3-y study conducted by Evans et al. (2009) to evaluate population response to the U.S. Department of Agriculture CP33 Habitat Buffers for Upland Wildlife program that reimburses private landowners for a share of costs to establish vegetation buffers on agricultural fields (Evans et al. 2009).

Data are from Fort Campbell Military Reservation, a 42,680-ha U.S. Department of Defense installation on the Tennessee–Kentucky state line. Study sites are in Montgomery and Stewart counties, Tennessee. Major vegetation types include hardwood forest, loblolly pine Pinus echinata plantations, native grass barrens, and agricultural fields. Because this site is a U.S. military reservation, those interested in additional details on the site should contact the authors for additional information.

In total, 110 point counts were randomly established at least 250 m apart. Points were located randomly using all available habitats on portions of the military reservation not having restricted access. Point counts were conducted by the same observer (E.D.M.) from 0500 to 1000 hours once during the last week of May through June during each year 2003–2006 and 2008–2009. Each count was 10 min in duration and recorded in three distinct time intervals. Bobwhite detections (by sight or sound) were placed into four distance radii: 0–25, 26–50, 51–100, and more than 100 m. Counts were not conducted during rain, heavy fog, or when winds were more than 6.5 km h⁻¹. Data analyses were performed with DISTANCE version 6.2 (Thomas et al. 2010) and SPSS version 17.0 (SPSS 2008). The raw data used in analysis are provided in the Supplemental Material (Table S1; http://dx.doi.org/10.3996/092010-JFWM-033.S1).

Of the 110 point counts, 85 had bobwhite detections and were analyzed in DISTANCE. Distance data were analyzed from point counts using the half-normal and uniform key functions and cosine, simple polynomial, and hermite polynomial expansion (Thomas et al. 2010). Akaike Information Criterion (AIC; Akaike 1974), weights, and goodness-of-fit χ² (α ≤ 0.05) were calculated for each model. The change in AIC was used to determine the most parsimonious models.

In total, 635 detections of bobwhites were made on 85 of the 110 points during the 6 y sampled. Observations were recorded in all bands: 0.3% (n  =  2) in the 0–25-m band, 3.5% (n  =  22) in the 25–50-m band, 22.2% (n  =  141) in the 51–100-m band, and 74% (n  =  470) in the more than 100-m band (Figure 1). No models were successfully fit with the key functions and expansions available in DISTANCE (Table 1). In all cases, although change in AIC values were acceptable (<2.0) based on the χ² test, the H₀  =  no difference in the observed frequency distribution and the distribution modeled by the key function and expansion was rejected (P < 0.05).

In this study, bobwhite detections using point count methods violated the assumptions of distance sampling. The assumptions of DISTANCE are 1) objects at the point are always detected, 2) objects are detected at their initial location before movement in response to the observer, 3) objects are correctly counted in the correct distance interval, 4) observers correctly identify the objects, 5) detections are independent events, and 6) the graph of the detection function has a shoulder around or near the point (Thomas et al. 2010).

Detection increased rather than decreased with distance from the observer (opposite to distance sampling theory; Thomas et al. 2010); there was no shoulder around or near the observation point. These data therefore violate the assumptions of distance sampling. It is likely that bobwhites are being influenced by observer presence by flushing away from the observer and not returning or not calling when close to the observation.

Northern bobwhite detection is difficult because bobwhite whistling, like all avian vocalization, is modulated in frequency, amplitude, or both (Valleru 2006). Bobwhites have auditory-localization ability, meaning that they can detect and thus project location into their vocalizations (Gatehouse and Shelton 1978; Gatehouse and Bailey 1986). Whistling males continually change position and direction of vocalization and when approached by other male(s) will “whisper” the “bobwhite” whistle (Stokes 1967). Thus, for humans, a loud whistle may be close or far depending on the singing bird's projection and position relative to the observer or its response to the observer. This has a profound effect upon an observer's ability to determine the distance from which a whistle is occurring in an environment where there are many competing sounds and environmental factors that can further modulate or filter sounds (Ramsey and Scott 1981; Simons et al. 2007). An observer simply cannot accurately place a bird within the proper distance band even if the bird is stationary. It is difficult for observers to accurately judge distances, especially beyond 100 m.

A lack of knowledge of the proportion of unmated to mated males within a target population is a problem with density estimation of bobwhite that cannot be reconciled with point counts and distance sampling during the breeding season. Because male calls are the cue being detected in a point count, the sex ratio of the population must be known to calculate population density (Buckland et al. 2001). We did not have an estimate of this multiplier for our study site and only report male density. Bobwhite females will mate with more than one male (Burger et al. 1995) so that sex ratio is critical to devising a multiplier for population estimation that truly reflects the target population density and trajectory. Knowing the sex ratio is even more critical because bobwhites are a hunted species, and population size after the hunting season is critical to predicting potential for reproduction.

Because avian conservation planning strives to track populations of multiple bird species, additional biases are introduced into the point count and distance sampling process. These biases include, but are not limited to, observer biases caused by hearing acuity, ability to estimate distance, expertise with bird vocalizations, distractions due to other species, attitude and condition of observers, and other observer-related factors (Faanes and Bystrak 1981; Scott et al. 1981). Differences in observers related to hearing ability are well documented (Emlen 1984; DeJong and Emlen 1985; Farnsworth et al. 2002). Older observers have poorer hearing acuity than do younger observers, and the result is missed vocalizations of some species and missed vocalizations at an observer-specific critical distance. Alldredge et al. (2008) found the assumption that observers could accurately map locations of aural detections was not met. Point counts are dependent on observer experience (Rappole et al. 1998) and are one of the two most important sources of bias in Breeding Bird Survey routes (DeMaso et al. 2002). Distractions from vocalizations of other bird species can bias counts of a particular species. Recent research has begun targeting one or a suite of species by using point-count protocol to increase efficiency (Alldredge et al. 2007). For example, the protocol for the mourning dove Zenaida macroura call-count survey does not permit counting of other species to avoid distractions (Sanders 2009). Multiple-observer and time-of-detection approaches as applied by Stanislav et al. (2010) and Riddle et al. (2010) do not compensate for problems with distance sampling found in this study.

The purpose of estimating density or abundance of bobwhites is to provide a measure of conservation success (MacKenzie and Nichols 2004). We suggest managers use caution in implementing avian point counts with distance sampling as a method of measuring bobwhite density. The use of distance sampling with point counts for many passerine bird species may be appropriate for estimating population density, but for bobwhite we recommended that managers continue to consider other alternatives. Bächler and Liechti (2007) suggested that alternative techniques be applied for the orphean warbler Sylvia hortensis. We suggest managers consider use of occupancy estimation and modeling (MacKenzie and Nichols 2004) as an alternative technique rather than abundance to monitor bobwhite populations. Furthermore, there is need to study observer influences and radius of audibility for bobwhite to better design monitoring approaches (Rusk et al. 2009).

Please note: The Journal of Fish and Wildlife Management is not responsible for the content or functionality of any supplemental material. Queries should be directed to the corresponding author for the article.

Table S1. Data used in analysis. Point  =  number that identifies individual counting point; year  =  year of count; distance  =  distance interval (meters) in which a bobwhite was counted.

Found at DOI: http://dx.doi.org/10.3996/092010-JFWM-033.S1 (21 KB XLSX).

We thank S. G. Somershoe for assistance with acquiring the data, discussions, and moral support and for reviewing an earlier draft of this manuscript. We thank the reviewers and Subject Editor of the Journal of Fish and Wildlife Management for valuable suggestions.

Any use of trade, product, website, or firm names in this publication is for descriptive purposes only and does not imply endorsement by the U.S. Government.