Aerial survey estimates of polar bears and their tracks in the Chukchi Sea

Abstract

Polar bears are of international conservation concern due to climate change but are difficult to study because of low densities and an expansive, circumpolar distribution. In a collaborative U.S.-Russian effort in spring of 2016, we used aerial surveys to detect and estimate the abundance of polar bears on sea ice in the Chukchi Sea. Our surveys used a combination of thermal imagery, digital photography, and human observations. Using spatio-temporal statistical models that related bear and track densities to physiographic and biological covariates (e.g., sea ice extent, resource selection functions derived from satellite tags), we predicted abundance and spatial distribution throughout our study area. Estimates of 2016 abundance ([Formula: see text]) ranged from 3,435 (95% CI: 2,300-5,131) to 5,444 (95% CI: 3,636-8,152) depending on the proportion of bears assumed to be missed on the transect line during Russian surveys (g(0)). Our point estimates are larger than, but of similar magnitude to, a recent estimate for the period 2008-2016 ([Formula: see text]; 95% CI 1,522-5,944) derived from an integrated population model applied to a slightly smaller area. Although a number of factors (e.g., equipment issues, differing platforms, low sample sizes, size of the study area relative to sampling effort) required us to make a number of assumptions to generate estimates, it establishes a useful lower bound for abundance, and suggests high spring polar bear densities on sea ice in Russian waters south of Wrangell Island. With future improvements, we suggest that springtime aerial surveys may represent a plausible avenue for studying abundance and distribution of polar bears and their prey over large, remote areas.

Fig 1. Chukchi Sea study area.

A map of the study area used in 2016 aerial surveys for seals and polar bears. Black lines represent aerial survey tracks, while small blue circles represent locations of bear tracks. Breaks in transect lines represent times where survey crew went “off effort” because of dense fog. Red triangles represent thermal detections of polar bear groups, and orange squares represent additional groups seen by human observers (U.S. and Russia) or in post hoc examination of photographs (Russia only). The ≈625km² grid cells used for abundance estimation appear in the background (beige lines). Land masses (gray) include Alaska, U.S.A. to the east and Russia, to the west. The blue shading represents the area associated with the Chukchi subpopulation of polar bears as determined by the polar bear specialist working group (PBSG), while yellow shading shows the area where a long term capture-recapture study of polar bears [8] was previously conducted. For an animation depicting survey effort and observations overlayed on sea ice as a function of survey day, see S1 Video.

Fig 2. Paired infrared and color imagery of a polar bear.

Color (right) and IR (left) imagery collected from 300 m during the 2016 aerial surveys from the U.S. platform containing a polar bear (zoomed inset). The IR image collected with the long wavelength infrared cooled camera (FLIR A6750sc SLS) provides a visual representation of apparent temperature in greyscale where darker shades are cool and lighter shades are warm. The paired color image, collected with the Prosilica GT6600c fitted with a 100 mm Zeiss lens, confirms that the heat signature detected in the IR image is from a polar bear.

Fig 3. Goodness-of-fit diagnostics.

Randomized quantile residuals (RQRs) for assessing goodness-of-fit for models fit to polar bear encounter data. RQRs should be uniformly distributed on (0,1) for a well-fitting model. Also presented are χ² test p-values to assess uniformity.

Fig 4. Spatio-temporal maps of sea ice and predicted polar bear distribution.

Remotely sensed sea ice concentration values (top row), estimated polar bear track index (middle row), and predictions of polar bear abundance at the beginning, middle, and end of 2016 aerial surveys of the eastern Chukchi Sea (g(0) = 0.8 scenario). The polar bear track index is an estimate of the proportion of photographs that would contain polar bear tracks had photographs been taken in all grid cells and on all days of the survey. Predicted abundance is calculated as ${\widehat{N}}_{s,t}=\widehat{N}{\widehat{\pi }}_{s,t}{\mu }_g$. Note that the scale of shading on abundance plots is nonlinear (i.e., low densities are visible as light blue and teal colors).

Fig 5. Covariate effects.

Estimated smooth effects of covariates on polar bear abundance (black line), together with 95% confidence intervals (grey shading). Note that distance from land, easting, and northing effects were standardized to have a mean of 1.0 prior to analysis. Polar bear tracks were modeled as a simple linear effect on abundance so do not appear here.