Integrating distance sampling with minimum counts to improve monitoring

Abstract

Minimum counts are commonly used to estimate population size and trend for wildlife conservation and management; however, the scope of inference based on such data is limited by untestable assumptions regarding the detection process. Alternative approaches, such as distance sampling, occupancy surveys, and repeated counts, can be employed to produce detection‐corrected estimates of population parameters. Unfortunately, these approaches can be more complicated and costly to implement, potentially limiting their use. We explored a conceptual framework linking datasets collected at different spatial scales under different survey designs, with the goal of improving inference. Specifically, we link landscape‐scale distance sampling surveys with local‐scale minimum counts in an integrated modeling framework to estimate mountain goat (Oreamnos americanus) abundance at both the local and regional scale in south‐central Alaska, USA, and provide an estimate of detection probability (i.e., sightability) for the minimum counts. Estimated sightability for the minimum count surveys was 0.67 (95% credible interval [CrI] = 0.52–0.83) and abundance for the entire area was 5,600 goats (CV = 9%), both in broad agreement with estimates from previous studies. Abundance estimates at the local scale (i.e., individual min. count unit) were reasonably precise (trueCV̅ = 18%), suggesting the integrated approach can increase the amount of information produced at both spatial scales by linking minimum count approaches with more rigorous survey designs. We propose that our integrated approach may be implemented in the context of a modified split‐panel monitoring design by altering survey protocols to include frequent minimum counts within local count units and intermittent but more rigorous survey designs with inference to the entire study area or population of interest. Doing so would provide estimates of abundance with appropriate measures of uncertainty at multiple spatial scales, thereby improving inference for population monitoring and management. © 2019 The Wildlife Society.

Comments on Integrating Distance Sampling and Minimum Count Data (Schmidt et al. 2019)

“…Perfect detection is an important assumption (Buckland et al 2015) on which the validity of the entire approach rests. Schmidt et al (2019Schmidt et al ( :1457 also state this assumption is critical and that bias occurs if it is not met.…”

“…Restrictive assumptions are needed to model minimum count data. Schmidt et al (2019Schmidt et al ( :1458 integrate the count data with distance sampling into a combined model with "the assumption that the same population is sampled" in both surveys. If the datasets are collected concurrently, the assumption that the same population is being sampled is valid.…”

“…Because the count areas are being searched over time to detect changes in population abundance, an assumption of no abundance change is unrealistic. Table 2 (Schmidt et al 2019) shows an augmented rotating panel sampling design in which a distance sampling survey is conducted every 5 years. All sites (count areas) are surveyed in years when a distance sampling survey is conducted and in the 4 years between distance sampling surveys, minimum counts are conducted in 25% of the sites on a rotating basis.…”

“…We argue the population estimate for this area is biased and not useful for goat management. Schmidt et al (2019Schmidt et al ( :1460 use the center-point of a transect to assign transect data to count areas. This methodology is mathematically incorrect when the transect lengths differ or when the transect traverses ≥2 count areas, which occurred at least once in 33 of the 35 goat count areas.…”

“…Estimated animal density is calculated by segment using the estimated distance sampling probabilities; an analogous method is used when creating density surface models from line transect data (Hedley andBuckland 2004, Miller et al 2013). The density segment estimates are weighted by the area searched along the transect segment to calculate a weighted mean density to estimate animal density for each count area, which is then used to construct the Schmidt et al (2019) count area sub-model. Misapplying the animal sightings to the wrong count area is analogous to putting a sample unit in the wrong strata when analyzing stratified sampling data.…”