Heterogeneity of locked-pasture snow conditions modulate habitat and movement choices of a facultative migrant, data archive

Habitat selection and movement are key mechanisms by which animals can respond to and potentially cope with highly variable and rapidly changing environmental conditions. Optimal responses likely vary, however, depending on the severity and scope of limiting conditions. We tested this hypothesis using a facultative migrant species, the Great Gray Owl (Strix nebulosa), which exhibits high inter- and intra-individual variation in the timing, direction, and distance of winter movements. Specifically, we evaluated whether espisodic, spatiotemporally variable “locked-pasture” snow conditions, which restrict access to subnivean food, prompted shifts in habitat selection or long-distance movements by owls. We GPS-tracked 42 owls across the annual cycle within the Greater Yellowstone Ecosystem, USA during 2017–2022. We used a novel ecological application of SnowModel, a snow evolution modeling system, to estimate fine-scale, physical snow properties likely to influence prey access. Variables included snow depth, snow crusts produced by wind, and ice crusts produced by melt-freeze and rain-on-snow events. Owls proximately avoided deeper snow and more severe, heterogeneously distributed wind crusts via local shifts in habitat selection. However, more widely distributed and persistent ice crusts elicited long-distance movements away from affected home ranges. Ultimately, the specific behavioral tactics employed varied with the severity, spatial extent, and duration of limiting environmental conditions. Our results provide a clear demonstration of behavioral flexibility in response to extremely limiting, highly variable environmental conditions. Such behavioral responses determine species distribution, with implications for community dynamics in spatiotemporally variable systems. Such understanding of responses to environmental variability is increasingly important given the scope of on-going global change. Methods Our research objective was to evaluate snow conditions in relation to Great Gray Owl movement behavior. We estimated snow depth and snow crust conditions across the study area (Greater Yellowstone Ecosystem) and period (1 September 2017 - 31 August 2022) using SnowModel. SnowModel is a spatiotemporally distributed snow-evolution modeling system that utilizes land cover, topography, meteorology, and in-situ observations of snow characteristics to simulate snowpack evolution across space and time. The modeling system is versatile in terms of spatiotemporal domain and resolution and can create nuanced, project-specific snow data products required to understand mechanistic snow-wildlife relationships. We modeled the evolution of snow conditions at two spatiotemporal resolutions to address our study objectives. First, we simulated snow characteristics at a three-hour time step and 30 m × 30 m spatial resolution to assess fine-scale habitat selection in response to local snow conditions. Second, we modeled snow evolution at a three-hour time step and 500 m × 500 m spatial resolution to assess probability of long-distance movements relative to broad-scale snow conditions. We used knowledge of the physical processes underlying snow crust formation to develop a comprehensive snow crust evolution submodel. Specifically, we implemented a physics-based model that simulated the spatiotemporal evolution of snow crusts associated with the three main phenomena that produce them: melt-freeze, rain-on-snow, and blowing and drifting snow associated with strong wind events. Melt-freeze crusts form via snowmelt occurring on the snow surface or within the snowpack, and then percolating down into sub-freezing snowpack layers where the snowmelt water refreezes; we call this a melt-freeze process or cycle. If the snowmelt water runs out of the bottom of the snowpack, a melt-freeze layer does not form. We identified snow crusts that formed during a melt-freeze cycle by calculating when snowmelt occurred within a cold snowpack (i.e., snowpack layer temperature < −1.0 °C) and accounted for subsequent freeze events after melting by identifying when snowmelt occurred followed by a snowpack layer temperature < −1.0 °C within the following 12 hr period. Rain-on-snow crusts form when rain precipitation falls on a cold (below freezing) snow surface or on a snowpack with cold internal snow layers, and the liquid freezes into an ice layer. Using the SnowModel output variables, we identified events during which rain (≥ 1.0 mm) fell on a cold snowpack (i.e., a snowpack layer with a depth ≥ 5.0 cm (based on minimum depth required to create subnivean habitat for small mammals) and temperature < −1.0 °C), and the liquid water froze in the surface layer, or layers below, and did not produce runoff. We indexed the severity of both melt-freeze and rain-on-snow crusts based on the amount of liquid water (rainfall or snowmelt, in millimeters) that reached a < −1.0 °C snowpack layer. Although ice crust severity was calculated simply as an index, the calculation is based on amount of liquid water (in millimeters) incorporated into the crust and, therefore, ice crust severity had units of 1.0 mm of liquid water frozen within the snowpack (e.g., an ice crust severity index of 10.0 meant 10.0 mm of liquid water had frozen). The reason we refer to this as an index is because the model is not performing a complete physics-driven energy and moisture balance; we have assumed that all that is required to freeze all the available liquid is the presence of cold snow (there could be instances where the snow was not cold enough to freeze all of the liquid that reached the cold layer). Melt-freeze and rain-on-snow crust indices were comparable in scale and formation process, so we combined melt-freeze and rain-on-snow events into a single variable: ice crust events. Our additional reasoning here was that only ice crust presence and magnitude was relevant to the focal species, not the precise processes that formed them. Wind crust formation occurs when high-speed winds (e.g., winds generally greater than 5 to 7 m s-1) interact with falling or blowing snow. Such events result in the mechanical breakage or tumbling of snow into small, sometimes sharp-edged crystals that can form strong, grain-to-grain bonds, through a process called sintering, once they stop moving. To model wind crust events, we calculated the change in snow density due to blowing snow based on the wind speed 2 m above the land surface. To index the severity of a wind crust event, we calculated the relative increase in snow density (again, for snowpacks with depth ≥ 5.0 cm based on required depth for subnivean habitat for small mammals) due to wind speed and amount of blowing snow. The severity indices for ice versus wind crusts are not directly comparable due to the inherent differences in how these crusts form in the natural system. For this reason, we did not combine the ice and wind crusts into a single, all-encompassing crust index. In our modeling system, once formed, both wind and ice crusts can persist, even as new snow falls and buries them below the surface. We calculated the formation and initial severity index of each crust (a crust event), and their cumulative severity index (a summed event severity index over time) and persistence (consecutive number of days that the crust persisted within the snowpack). In our ice and wind crust model, crusts disappear from a snowpack when it becomes isothermal at 0.0 °C and liquid snowmelt water reaches the ground (typically in the spring, or in low elevations, periodically throughout the winter). Once runoff occurred, we reset the cumulative variables (cumulative severity index, persistence) to zero, to represent this process. Additionally, we collected GPS locations for tagged, adult Strix nebulosa (Great Gray Owls) (n = 42) in northwestern Wyoming between 2017-2022. We used a used-available framework and integrated step-selection analysis to evaluate step selection by owls in relation to snow conditions. A step consists of both a starting and ending location between a set time interval. This analysis assumes that, for each observed step, alternate available steps exist that the animal could have selected. We subsampled Great Gray Owl GPS location data to include the core snow period defined as the average dates of the longest period during which snow cover was consistent (15 September–15 April) and one location per day, such that successive locations were 24 hr apart. For each observed step, we genearted generated five corresponding random steps for each observed step, drawn from theoretical distributions of average step lengths and turning angles for that specific individual (Weibull distribution and Von Mises distribution for step-length and turning angle, respectively). At the endpoint of each step, we extracted values of snow depth; wind crust conditions (severity, cumulative severity, and persistence); and ice crust conditions (severity, cumulative severity, and persistence). To evaluate the snow conditions that precipitated migratory movements by owls, we identified long-distance movement events by S. nebulosa during the core snow period (15 September – 15 April). We used Net Squared Displacement analysis, which characterizes movement trajectories by calculating the squared Euclidean distance between a starting location and subsequent locations. We also visually inspected GPS location data to confirm that long-distance movement events met specific criteria. We identified a migratory movement event as a significant movement away from a discrete breeding or winter range. A significant movement was defined as a movement of 8 km or farther (based on twice the diameter of the mean overall winter range area (19.5 km2, based on 99% Kernel Density Estimates (KDE)). We defined a breeding home range based on a 95% KDE using locations between April–September. A discrete winter home range was defined as an area < 3.7 km in diameter (based on mean winter home range size of 13.4 km2 95% KDE) in which an owl settled for a minimum of 1 week (between 1 October–30 March). We excluded movements away from discrete winter ranges in which owls returned to their breeding range, because we could not ascertain whether factors other than snow conditions (e.g., acquisition and defense of a territory and mate) influenced decisions. We subsampled all GPS locations to one location per day per individual and excluded movement sequences with > 5 days of missing location data and/or < 6 months of GPS location data total. For each long-distance movement event, we identified the onset of risk as the point at which the owl first settled in either a discrete breeding or winter range (settlement decision), and included all locations in which the owl remained settled (settlement) up until the point at which the owl departed the settled range to undertake a long-distance movement (departure event). For all locations beginning with the settlement decision through the departure event, we extracted values of snow depth and crust severity, cumulative crust severity, and crust persistence for both ice and wind crusts. We used Cox proportional hazards analysis to evaluate whether snow conditions influenced probability of departure by Great Gray Owls.