Peter J. Mahoney, Glen E. Liston, Scott LaPoint, Eliezer Gurarie, Buck Mangipane, Adam G. Wells, Todd J. Brinkman, Jan U. H. Eitel, Mark Hebblewhite, Anne W. Nolin, Natalie Boelman, Laura R. Prugh

Winters are limiting for many terrestrial animals due to energy deficits brought on by resource scarcity and the increased metabolic costs of thermoregulation and traveling through snow. A better understanding of how animals respond to snow conditions is needed to predict the impacts of climate change on wildlife. We compared the performance of remotely sensed and modeled snow products as predictors of winter movements at multiple spatial and temporal scales using a data set of 20,544 locations from 30 GPS-collared Dall sheep (Ovis dalli dalli) in Lake Clark National Park and Preserve, Alaska, USA from 2005 to 2008. We used daily 500-m MODIS normalized difference snow index (NDSI), and multi-resolution snow depth and density outputs from a snowpack evolution model (SnowModel), as covariates in step selection functions. We predicted that modeled snow depth would perform best across all scales of selection due to more informative spatiotemporal variation and relevance to animal movement. Our results indicated that adding any of the evaluated snow metrics substantially improved model performance and helped characterize winter Dall sheep movements. As expected, SnowModel-simulated snow depth outperformed NDSI at fine-to-moderate scales of selection (step scales < 112 h). At the finest scale, Dall sheep selected for snow depths below mean chest height (<54 cm) when in low-density snows (100 kg/m3 ), which may have facilitated access to ground forage and reduced energy expenditure while traveling. However, sheep selected for higher snow densities (>300 kg/m3 ) at snow depths above chest height, which likely further reduced energy expenditure by limiting hoof penetration in deeper snows. At moderate-to-coarse scales (112-896 h step scales), however, NDSI was the best-performing snow covariate. Thus, the use of publicly available, remotely sensed, snow cover products can substantially improve models of animal movement, particularly in cases where movement distances exceed the MODIS 500-m grid threshold. However, remote sensing products may require substantial data thinning due to cloud cover, potentially limiting its power in cases where complex models are necessary. Snowpack evolution models such as SnowModel offer users increased flexibility at the expense of added complexity, but can provide critical insights into fine-scale responses to rapidly changing snow properties.