Main content area

High variation in camera trap-model sensitivity for surveying mammal species in northern Australia

Heiniger, Jaime, Gillespie, Graeme
Wildlife research 2018 v.45 no.7 pp. 578-585
cameras, field experimentation, models, probability, small mammals, surveys, wildlife, Northern Territory
Context. The use of camera traps as a wildlife survey tool has rapidly increased, and understanding the strengths and weaknesses of the technology is imperative to assess the degree to which research objectives are met. Aims. We evaluated the differences in performance among three Reconyx camera-trap models, namely, a custom-modified high-sensitivity PC850, and unmodified PC850 and HC550. Methods. We undertook a controlled field trial to compare the performance of the three models on Groote Eylandt, Northern Territory, by observing the ability of each model to detect the removal of a bait by native mammals. We compared variation in detecting the known event, trigger numbers, proportion of false triggers and the difference in detection probability of small to medium-sized mammals. Key results. The high-sensitivity PC850 model detected bait take 75% of the time, as opposed to 33.3% and 20% for the respective unmodified models. The high-sensitivity model also increased the detection probability of the smallest mammal species from 0.09 to 0.34. However, there was no significant difference in detection probability for medium-sized mammals. Conclusions. Despite the three Reconyx camera models having similar manufacturer-listed specifications, they varied substantially in their performance. The high-sensitivity model vastly improved the detection of known events and the detection probability of small mammals in northern Australia. Implications. Failure to consider variation in camera-trap performance can lead to inaccurate conclusions when multiple camera models are used. Consequently, researchers should carefully consider the parameters and capabilities of camera models in study designs. Camera models and their configurations should be reported in methods, and variation in detection probabilities among different models and configurations should be incorporated into analyses.