Submitted by editor on 3 April 2018. Get the paper!
Photo: Karen Lone, Norsk Polarinstitutt​

By Karen Lone, Benjamin Merkel, Christian Lydersen, Kit M. Kovacs, Jon Aars

Polar bears use sea ice as a primary hunting habitat. Climate change is causing major declines in this unique habitat. So - how is this altering the distribution and ecology of this iconic Arctic species?

We have investigated this question using resource selection functions (RSFs) to model polar bear habitat use in the Barents Sea area; this region is home to one of the 19 subpopulations of polar bears globally. Using data from 226 female polar bears, spanning the period 1991-2015, we created seasonal models that predicted well across spatial, temporal and random subsets of the data. These “triple model selection criteria” were proposed by Wiens et al. (2008), who argued that “a RSF model’s utility is its ability to predict”.

The large, long-term dataset underlying the RSF models was very advantageous in dealing with a highly variable system. The Barents Sea can experience dramatic differences in the sea ice cover between years, as well as the major retreat of the sea ice that has been taking place over the last three decades (Laidre et al. 2015). What emerged in our models was therefore which habitat characteristics polar bears selected with some consistency over the variable sea ice conditions during the last 25-year time period, across the study area. The primary finding was that the polar bears use the transition zone between dense pack ice and open water, a region commonly referred to as the Marginal Ice Zone (MIZ).

Can we say something about how the polar bears’ habitat is changing? Yes, we can. We can look at how the distribution of optimal habitat has changed in the recent past, and make predictions about the future. The findings are very much in line with what Durner et al. (2009) suggested for this area using their global polar bear habitat model. The duration and extent (habitat-day pixels) of optimal polar bear habitat has declined by 27% over the 25-year study period. During this same time, there has been a remarkable northward shift in where their preferred habitat occurs. The MIZ is a dynamic entity that moves over a matter of days, but also over years, and it is occurring farther and farther north as sea ice loss continues as a result of global warming.

Figure 1: Number of habitat-days (left panel) at each location (pixel), and change in these over time (middle and right panel).​

Predicting for a future with sea ice conditions that have never before occurred in human history obviously involves quite extensive extrapolation. The quality of the habitat may suddenly improve, much more likely, deteriorate if seals, the main prey of polar bears, alter their behaviour in response to the sea ice changes, or decline markedly because of poor breeding conditions for them. But, with some certainty, we can say that moving between denning habitats on land (pregnant females in the area need to reach islands where they give birth to cubs mid-winter in snow drift dens) and hunting habitats in the drifting sea ice will become more difficult as the two are becoming increasingly spatially separated. This single factor will result in increased transport costs for the bears and more energy spent swimming. It might induce new migratory space use patterns. Exactly how a continued loss of drifting sea ice will effect the population dynamics of polar bears remains to be seen, but the Barents Sea subpopulation will inevitably face conditions that are more challenging. Declines in their primary hunting habitat will undoubtedly have negative consequences.

Photo: Kit Kovacs and Christian Lydersen, Norsk Polarinstitutt​


Durner et al. 2009. Predicting 21st-century polar bear habitat distribution from global climate models. – Ecol. Monogr. 79: 25-58.

Laidre et al. 2015. Arctic marine mammal population status, sea ice habitat loss, and conservation recommendations for the 21st century. – Conserv. Biol. 29: 724–737.

Wiens et al. 2008. Three way k-fold cross-validation of resource selection functions. – Ecol. Model. 212: 244–255.