Estimating Detection Probabilities of Waterfowl Broods From Ground‐Based Surveys

Abstract

Brood:pair ratios could provide an economical method for assessing spatial or temporal variation in waterfowl productivity, but such estimators are severely biased by incomplete detection of broods. We conducted 3 sequential counts of 1,357 waterfowl broods in northeastern North Dakota, USA, and used closed‐population mark‐recapture models to estimate total brood abundance while controlling for variation in detection probabilities (p). Blue‐winged teal (Anas discors) broods had the lowest average detection probability (p = 0.305), whereas diving‐duck broods had the highest average detectability (p = 0.571). Detection was generally highest in morning or evening, but temporal patterns varied among species and there was no survey window that maximized detection probabilities for all species. Detection probabilities averaged 0.108 (SD = 0.056) higher for an experienced observer versus an inexperienced observer. Detection probabilities were 0.044 higher for roadside versus walk‐up surveys and increased with increasing brood size, total brood abundance, survey date, wind speed, temperature, cloud cover, and amount of time spent surveying each wetland. Detection probabilities declined with increasing wetland size and amount of tall peripheral vegetation. Our mark‐recapture results indicated that a traditional unreplicated brood survey would have missed 67.5% of estimated broods, summed over all species. Use of closed‐population mark‐recapture techniques provided an effective method for reducing this bias and identifying and quantifying factors that reduce detection probabilities of waterfowl broods. We recommend that future brood surveys incorporate 2 or 3 temporally segregated replicate counts to allow for formal estimation of detection probabilities.

On the efficiency of using song playback during call count surveys of Red-legged partridges (Alectoris rufa)

“…The main findings of this paper are that (a) the probability of detecting singing males within the framework of traditional aural point-count protocols was low compared to other bird species (Eraud et al 2007;Gonzalo-Turpin et al 2008;Pagano and Arnold 2009) and (b) the additional use of playback strongly increased detectability, therefore improving the accuracy of abundance estimates. We observed differences in detectability between the 2 years, with higher values recorded in 2009.…”

Habitat associations and survey effort for shrubland birds in an urban pine barrens preserve

“…This was the most parameterized (about four sample points per model parameter) of the six models. We also modeled the influence of habitat type (p(HABITAT)), observer identity (p(OBSERVER)), and time of day (p(STARTTIME)) on detection probability, as all of these factors can influence the potential detectability of numerous bird species (Chandler et al, 2009;Darrah and Krementz, 2009;Gonzalo-Turpin et al, 2008;King et al, 2009a,b;Mitchell and Donovan, 2008;Pagano and Arnold, 2009;Ruiz-Gutiérrez et al, 2010;Skirven, 1981). We converted habitat cover and point count start times into standard normal deviates (z-scores) to help facilitate convergence of the numerical optimization algorithm, and modeled observers as categories (observer 1 vs. 2).…”

“…Royle and Nichols (2003) suggested that 10 surveys can produce good model abundance estimates even for bird species with low detection probability. An effort of three surveys, as suggested by the Northeast Shrubland Bird Workgroup and elsewhere (Field et al, 2005;MacKenzie and Royle, 2005;Pagano and Arnold, 2009;Tyre et al, 2003) as a general rule, may be insufficient for shrubland birds in pine barrens. Note, however, that because detectability tends to vary by habitat type or structural characteristics (Chandler et al, 2009;Chace et al, 2009;Darrah and Krementz, 2009;GonzaloTurpin et al, 2008;Mitchell and Donovan, 2008;Ruiz-Gutiérrez et al, 2010), sample points located in habitat where a species is more detectable may require fewer surveys.…”

“…Using two observers for 10-min point counts, assuming ∼5-min travel time between points, and restricting surveys to the ideal time for detection (05:00-08:00 hr), it should take between 2 and 3 days to cover all points on a sampling occasion. Using more than two observers is allowable, but given that we and others (e.g., Chandler et al, 2009;Darrah and Krementz, 2009;King et al, 2009a;Pagano and Arnold, 2009) have documented clear differences in observers' ability to detect birds, it is important to model detection probability as a function of observer heterogeneity. Alternatively, we found that reducing sample size to as few as ∼30 points (∼2% of preserve area) may retain 0.05-0.10 SE on occupancy estimates, at least if 10 or more surveys are conducted.…”

“…Because detection of animals often varies among habitat types and structural characteristics (Chandler et al, 2009;Chace et al, 2009;Darrah and Krementz, 2009;Gonzalo-Turpin et al, 2008;Mitchell and Donovan, 2008;Ruiz-Gutiérrez et al, 2010), neglecting to correct for detection probability may lead to erroneous conclusions based on a species' detectability in different habitats rather than its actual habitat preferences. Recent bird studies have incorporated an occupancy-detection modeling approach to rigorously test hypotheses about habitat associations (Betts et al, 2008;DeWan et al, 2009;Hennemen and Andersen, 2009;Kroll et al, 2007;Pagano and Arnold, 2009;Richmond et al, 2008), but such analyses have not been done for shrubland birds in pine barrens. Grand and Cushman (2003) suggested that species-level habitat models of birds commonly associated with pine barrens will help managers create the appropriate mix of pine barrens successional communities (i.e., tree-vs. shrub-dominated habitat).…”