Testing the Accuracy of Aerial Surveys for Large Mammals: An Experiment with African Savanna Elephants (Loxodonta africana)

Abstract

Accurate counts of animals are critical for prioritizing conservation efforts. Past research, however, suggests that observers on aerial surveys may fail to detect all individuals of the target species present in the survey area. Such errors could bias population estimates low and confound trend estimation. We used two approaches to assess the accuracy of aerial surveys for African savanna elephants (Loxodonta africana) in northern Botswana. First, we used double-observer sampling, in which two observers make observations on the same herds, to estimate detectability of elephants and determine what variables affect it. Second, we compared total counts, a complete survey of the entire study area, against sample counts, in which only a portion of the study area is sampled. Total counts are often considered a complete census, so comparing total counts against sample counts can help to determine if sample counts are underestimating elephant numbers. We estimated that observers detected only 76% ± SE of 2% of elephant herds and 87 ± 1% of individual elephants present in survey strips. Detectability increased strongly with elephant herd size. Out of the four observers used in total, one observer had a lower detection probability than the other three, and detectability was higher in the rear row of seats than the front. The habitat immediately adjacent to animals also affected detectability, with detection more likely in more open habitats. Total counts were not statistically distinguishable from sample counts. Because, however, the double-observer samples revealed that observers missed 13% of elephants, we conclude that total counts may be undercounting elephants as well. These results suggest that elephant population estimates from both sample and total counts are biased low. Because factors such as observer and habitat affected detectability of elephants, comparisons of elephant populations across time or space may be confounded. We encourage survey teams to incorporate detectability analysis in all aerial surveys for mammals.

Using species traits to predict detectability of animals on aerial surveys

“…Observer effects were not the focus of our study, but excluding them could lead to inaccurate predictions. A detailed study of elephants (Schlossberg et al 2016) and exploratory analyses of the multi-species data set used in this study revealed that three covariates related to observers influenced detectability. First, one observer had a higher detection probability than the other three observers.…”

“…, Graham and Bell , Ransom , Schlossberg et al. ). Research has also shown that detectability of mammals decreases as vegetation density increases (Bayliss and Yeomans , Anderson and Lindzey , Cogan and Diefenbach ).…”

“…, Ransom , Schlossberg et al. ). In this study, one of the two most experienced observers had a higher detection probability than the other three observers.…”

“…Though our goal was to model species traits, past research suggested that two herd-level traits, the number of animals in a herd and the type of vegetation around the herd, would also affect detectability. Herd size is consistently the most important factor affecting detectability of group-forming species on aerial surveys (Samuel et al 1987, Graham and Bell 1989, Ransom 2012, Schlossberg et al 2016. Research has also shown that detectability of mammals decreases as vegetation density increases (Bayliss and Yeomans 1989, Anderson and Lindzey 1996, Cogan and Diefenbach 1998.…”

“…First, one observer had a higher detection probability than the other three observers. Second, detection probabilities were higher in the rear row of seats than in the front row of seats in the Airvan (see Schlossberg et al 2016). Third, detectability increased with a species' abundance, likely due to observers' search image improving as they repeatedly observe a species.…”