The world's polar bear (Ursus maritimus) population occurs only where Arctic sea ice is predominant during the annual cycle. The veneer of ice over marine waters allows polar bears to hunt the ice-dependent pinnipeds which they depend on for their survival. Since 1979 the extent and volume of sea ice has diminished, displacing polar bears from prey-rich waters and reducing the quality of remaining ice habitat. To conserve polar bears in a warming Arctic it is necessary to understand what sea ice types are preferred, how preferred habitats have and will changed, and to determine whether the movement behavior of polar bear has been altered to compensate for sea ice declines. To increase our understanding of polar bear response to a changing Arctic and inform decisions for polar bear conservation I used satellite telemetry data from adult female polar bears to develop coarse-grain, polar basin-wide resource selection functions (RSF) from passive microwave (PM) imagery of sea ice concentration and extent, and fine-grain Beaufort Sea RSFs from National Ice Center (NIC) data of sea ice age and composition. I used the coarse-grain RSFs to measure changes in optimal habitats between 1985-1995 and 1996-2006 for the polar basin. I then applied those RSFs to 10 general circulation models (GCM) predictions of 21 st century sea ice to quantify changes in polar bear preferred sea ice habitat. I used the fine-grain RSFs to identify sea ice structure and composition preferred by polar bears in the Beaufort Sea. I then compared the rate of habitat change indicated by the fine-grain RSFs to coarse-grain RSFs, GCM predictions, and sea ice extent and thickness. Finally, I examined the behavioral response of polar bears to compensate for a substrate that was constantly moving under their feet and how this response differed between the Beaufort and the Chukchi seas and from a period of relatively stable ice to a period of diminished sea ice. Within the polar basin and from 1985-1995 to 1996-2006 optimal sea ice habitats declined in all regions over contented shelf waters except for regions adjacent to northern Greenland and the northern shores of the Canadian Archipelago. When applied to GCMs to project 21 st century habitat conditions, habitat loss continued in the southern seas of the polar basin, from the Beaufort Sea in Alaska and Canada, across northern Russia, to the Barents Sea. Habitat loss was minimal, and sometimes improved, along the Arctic Ocean shores of Banks Island to northern Greenland. Optimal polar bear habitat declined most during summer; from an observed 1.0 million km 2 in 1985-1995 to a projected multi-model average of 0.32 million km2 in 2090-2099 (-68% change). Loss of winter habitat was less; from 1.7 million km 2 in 1985-1995 to 1.4 million km2 in 2090-2099 (-17% change). Despite projected declines of future habitat conditions actual changes may be occurring faster than predicted as observed habitat loss during 1985-2006 was greater than the loss predicted by GCM hindcasts. The fine-grain RSFs for the Beaufort Sea for 1999-2012 showed that polar bears selected pack ice in shallow continental shelf waters near land fast ice, ice edges, and young sea-ice during winter and spring. From breakup to freeze-up, when sea ice was often over waters >2500 m deep, polar bears selected high-concentration ice that minimized their distance from the continental shelf. In all seasons except summer the distribution of habitats preferred by polar bears coincided with the expected distribution of seal prey. During the years of this study both NIC- and PM- based RSFs showed a decline in habitat quality during breakup, summer and freeze-up. But habitat declines outpaced those predicted by GCMs, were similar to declines in overall ice extent, and lower than sea ice loss indicated by ice thickness trends. NIC-based RSFs provided an improvement for understanding the ecological relationship between polar bears and sea ice and that, despite overt declines in the quantity of sea ice, some preferred habitats persisted in every season throughout this study. Understanding changes in preferred sea ice habitats does not by itself explain energetic consequences of diminishing sea ice on polar bears. Changes in sea ice drift rates from thinning ice provides a link between the environment and the costs of locomotion by polar bears. In Chapter 3 I compared daily movement steps by polar bears relative to sea ice drift rates between the Beaufort and the Chukchi seas and across three periods with distinct sea ice characteristics: 1985-1995, 1996-2006 and 2007-2013. I used a continuous-time correlated random walk model to regularize radio collar locations to one every 24 hours. By accounting for daily sea ice drift I was able to derive ice-corrected "true" polar bear movements. Sea ice drift rates increased sequentially across the time periods and were greater in the Beaufort Sea than in the Chukchi Sea but the azimuth of ice drift remained unchanged across periods. Rates of bear movements and collars displacements were similar but their azimuths differed, suggesting that collar locations are not indicative of animal movements on a moving substrate. In either region bear movements were eastward and opposite or perpendicular to ice drift and distances between observations increased from 1985-1995 to 2007-2013. Polar bears in the Beaufort Sea responded to changes in ice drift directionally, while bears in the Chukchi Sea compensated by increasing the distance traveled between locations. As they moved from one location to the next, polar bears showed greater selection for higher quality habitat in 2007-2013 than in 1985-1995, suggesting that habitat quality had declined due to overt reductions in sea ice and increased energetic costs for polar bears to seek and remain in the best habitat. Optimal polar bear habitat in the polar basin will decline throughout the 21st century. Trends in habitat as predicted by observational data suggest that predictions of habitat decline from GCMs may not be realistically severe, i.e., habitat loss is occurring faster than predicted by models. However, habitat loss is not uniform throughout the Arctic as high latitude regions adjacent to north Greenland and the northern Canadian Archipelago are likely to retain sufficient sea ice habitat to serves as a refugia for a viable, albeit greatly reduced, polar bear population. Additionally, fine grain models indicate that some optimal habitat may persist even in an Arctic with reduced sea ice. That remnant habitat will be susceptible to an increasingly thinning ice cover and loss through stochastic events. In addition to overt declines in sea ice the energetic costs to polar bear to seek and occupy optimal sea ice habitat will likely increase as thinning sea ice becomes increasingly mobile due to the actions of winds and currents.