Misidentification error associated with classifications of white‐tailed deer images

Abstract

Camera traps are widely used to monitor wildlife, with important management decisions often relying on interpretation of these data. Animal misidentifications are known to be an important source of error in wildlife surveys that require the identification of unique individuals from camera-trap data; however, the practice of broadly classifying animal images according to sex or age has received less critical attention despite the significant potential for misidentification error under certain circumstances. From 19 January to 1 April 2017, we solicited a group of 726 participants, consisting of both wildlife professionals and nonprofessionals from across the United States, to take an online survey that tested their ability to classify images of known white-tailed deer (Odocoileus virginianus) according to sex and age. Our goal was to determine the relative influence of tested observer (i.e., experience and familiarity with classifying deer images) and image-based factors (i.e., distance of deer from camera, day vs. night image) on accuracy of deer classifications. Our results indicated that respondents that were wildlife biologists and those with greater levels of experience viewing deer images were more accurate than others when classifying posthunting season images of deer as adult male, adult female, or fawn. However, the sex-age group of the deer was the most influential predictor of classification reliability, with branched-antlered adult males being classified more accurately by all respondent groups than were adult females and fawns. Our findings emphasized that animal misidentifications may be an important source of survey error not only when identifying unique individuals, but also under any circumstance where comparative groups lack definitive traits. We suggest that those using camera traps to evaluate wildlife populations should select survey periods that maximize differences among classification groups, when possible, and develop species-specific image training for observers to improve the reliability of results. Further, population demographics should be considered when evaluating the overall reliability of survey results for species where classification accuracy varies among sex-age groups.

Inter-observer variance and agreement of wildlife information extracted from camera trap images

“…Most identification mistakes occurred between species with least phenotypic distinctiveness, for instance when observers confused Plains and Hartmann's mountain zebras. This finding is consistent with other camera trap studies showing that accurate mammal classifications are difficult to obtain where a range of similar-looking species occur sympatrically (Gooliaff and Hodges 2018), or individuals of a particular species exhibit little distinctiveness (Güthlin et al 2014;Newbolt and Ditchkoff 2019). Following our study, observers commented that animal orientation in the image frame (broad side vs. frontal vs. rear views) as well as 'crowdedness', i.e.…”

“…Previous research suggests not, showing that inter-observer variance can greatly affect the reliability of results obtained when multiple human observers are expected to perform the same task. Studies found considerable inconsistencies in the information obtained, be it during the identification of mammalian species or unique individuals from camera trap footage (Gooliaff and Hodges 2018;Johansson et al 2020), or the classification of population characteristics (Newbolt and Ditchkoff 2019). It is also clear that different factors such as camera type (Randler and Kalb 2018) and observer experience can influence the results (Burns et al 2018;Katrak-Adefowora Fig.…”

Reproduction of white-tailed deer in a seasonally dry tropical forest of Costa Rica: a test of aseasonality

“…A deer was un-categorized if only a portion of the deer was visible, such as a tail or leg. Three of us (AMS, TKF, and VHM) identified all photos; we trained together and repeatedly interacted on questionable categorizations to minimize misclassifications (Newbolt and Ditchkoff 2019).…”

Influence of camera model and alignment on the performance of paired camera stations

“…We visually leveled each camera, and tested FOV using a standard visibility marker flag (Hofmeester et al 2017, Wearn and Glover-Kapfer 2017, Moll et al 2020; none of the camera models detected our visibility marker flag at >3 m. To evaluate potential effects of aligned cameras on station performance, we placed like model cameras directly across from one another, and perpendicular to maintained paths and trails (Figure 1). To examine the performance of stations with staggered cameras, we faced the cameras directly across a maintained path or trail and offset them by 4.6 m. Stations were not baited (i.e., to minimize bias [Newbolt et al 2017, Fidino et al 2020) and set up to capture F I G U R E 1 Diagram of a 2 paired camera station; alignment options included alternating camera models between aligned camera stations (4.6 m across from each other and perpendicular to the trail) and staggered camera stations (4.6 m across from each other and perpendicular to the trail and offset by 4.6 m) evaluated at Kibbe Station, Warsaw, Illinois, USA, summer 2018. Paired camera stations were separated by ≥9 m to ensure independence (i.e., flash from one station would not impact images at subsequent stations) between successive stations, and we alternated camera station models across treatments to minimize environmental bias.…”