Abstract 289 – Using anomaly detection with radio-collar activity data to determine the presence and timing of grizzly bear parturition

Lori Roberts, Cecily M. Costello, Milan A. Vinks, Daniel D. Bjornlie, Matthew D.
Cameron, Justin G. Clapp, Mark A. Haroldson, Grant V. Hilderbrand, Kyle Joly, Wayne Kasworm,
Jeremy Nicholson, Tom Radandt, Mathew S. Sorum, Justin Tiesberg, Frank T. van Manen

Documenting natality of radio collared grizzly bears (Ursus arctos) is a key component of many
population monitoring programs. Because grizzly bears give birth during hibernation,
observation of litters is not possible until den exit and detection of cubs can be compromised
by low sightability. Using activity data from dual-axis motion sensors, previous researchers
developed methods to confirm parturition based on characteristic changes in activity between
pregnancy and postpartum periods. Unable to replicate their results with data from 3-axis
accelerometers, we developed an alternative technique to detect births by the presence of
spikes in activity likely triggered by prolonged postpartum licking by mothers. Using a test
sample of daily activity counts during 25 Dec–7 Mar (n = 22 known parturient females), we
applied the R routine anomolize and developed presence criteria. To test performance, we
applied the technique to a blind sample of time-series obtained from females in 4 populations
in interior North America (n = 296). We assigned predicted status and compared assignments
to observed status at first visual observation. The true positive rate was 94% (n = 47 observed
parturient females). The false positive rate was 14% (n = 65 females observed with older
offspring). Births were predicted for 47% of females that were never observed (n = 49), 21% of
females observed without offspring (n = 114), and 10% of females considered too young to
reproduce (n = 21). Dates of predicted births varied from 27 Dec to 28 Feb. Our anomaly
detection technique was successful at estimating parturition events, and despite some error,
indicated that a number of litters were born but not observed, likely because of mortality in the
den or shortly after den exit. This technique may provide an additional tool for supplementing
visual observations for natality monitoring and population modeling, provided that potential
biases stemming from increased detection rates are considered.