Enhancements to population monitoring of Yellowstone grizzly bears

Abstract

In the Greater Yellowstone Ecosystem, counts of female grizzly bears (Ursus arctos) with cubs-of-the-year (females with cubs) from systematic aerial surveys and opportunistic ground sightings are combined with demographic data to derive annual population estimates. We addressed 2 limitations to the monitoring approach. As part of a rule set, a conservative distance of >30 km currently is used as a threshold to assign sightings to unique females with cubs, resulting in underestimation bias. Using telemetry locations of females with cubs collected during 1997–2019, we created 1,000 data sets for each of 5 levels of simulated number of females with cubs, simulated sightings by selecting among these locations, and evaluated the classification performance of alternative distance criteria (12–30 km). Under all scenarios, 12–16-km criteria maximized classification performance and minimized estimation bias; the 16-km criterion was optimal for current conditions and sampling efforts. Our second objective was to test generalized additive models (GAMs) as a flexible trend analysis technique. We simulated 1,000 time series for each of 10 scenarios (10, 15, and 20% decline over periods of 5, 10, and 15 yrs, plus stability), applied GAMs, and assessed metrics associated with the posterior distribution of the instantaneous rate of change. We detected declines among >99.6% of replicates under the 15 and 20% decline scenarios and in 84.7–94.7% of replicates under the 10% decline scenario. From decline onset to first detection, periods ranged from 3.7 (20% decline over 5 yrs) to 11.1 (10% decline over 15 yrs), with 3.9–8.8 years mean duration of detection events. The GAM approach allows detection of directional changes in population trend, including early warning metrics, and stabilization after such changes. Retrospective application of the 16-km distance criterion and GAMs resulted in higher population estimates and growth rates than are reported using current methods.