Model suitable and diagnostics
We fit our Bayesian state-space model, consisting of both movement and observation process components, using Markov chain Monte Carlo (MCMC) methods. We assigned uninformative priors whenever there were no a priori expectations about the relationships of the model parameters and the movement behavior states (see Appendix S1). letter,t) within each iteration of the MCMC algorithm (see Appendix S2). We initialized six chains from random starting values drawn from the prior distributions. After initial pilot tuning and burn-in, we ran the chains until the univariate Gelman–Rubin potential scale reduction factor (PSRF) for monitored parameters was <1.2 and effective sample sizes exceeded 4000. This required about 6 million iterations per chain, where 1 million iterations for each parallel chain required about 100 h using six cores of a 3.7-GHz processor. Data and code can be found in Data S1.
Performance
Based on posterior settings to own county projects, we discover these bearded seals invested 70% of its deployment episodes from the benthic foraging county (B). They hauled from ice (state We) and you will rested within water (condition S) within the more or less equivalent dimensions, however, there’s a very clear regular part into the craft finances certainly one of the fresh new hauled on ice (I), benthic foraging (B), and transit (T) claims one to coincided to the june haven and you will winter months progress off ocean ice throughout the Arctic (Table dos). All the seals exhibited rigorous benthic foraging along side continental cabinets from the new Chukchi otherwise Beaufort seas given that ocean freeze receded northward for the the summer months. I discover little proof this type of frost-related seals pulling from home, even though near the coastline for the mostly freeze-totally free period during summer. Surprisingly, a grownup men briefly gone north of your Beaufort seaside shelf inside later and you can was predicted to own hauled on the falling out in clumps ocean ice border ranging from bouts of middle-drinking water foraging on deeper seas of the Canada Basin (Fig. 2). We offer a detailed animation away from predicted movements and you will state projects for everybody seals in line with bathymetry and you may sea ice security inside the Appendix S3.
- There were seasonal differences in hobby budgets ranging from “summer” (out-of tagging when you look at the late June and you can early July to 30 Sep), “autumn” (step one October to 29 December), and you can “winter” (1 January up until tag losses anywhere between February and you will April) you to coincided on the southern area get better from winter months ocean freeze into the the latest Cold. Interest finances descriptions have been calculated according to the posterior mode away from date spent within the for each and every state having down (LCI) and you will upper (UCI) legitimate times predicated on 95% higher rear density. Private secure hobby spending plans are included in Appendix S4.
As expected, the resting states (I, S, and L) San Antonio casual hookup exhibited shorter step lengths, no directional persistence, and lower proportions of each time step diving >4 m below the surface based on state-specific parameter estimates (Fig. 3; see Appendix S4 for posterior sumeters). Both the mid-water foraging (M) and transit (T) states exhibited longer step lengths, lower proportions of dry time, and higher proportions of diving >4 m below the surface. Only the transit state exhibited strong directional persistence, but state M exhibited moderate directional persistence for some individuals (Fig. 3). The benthic foraging (B) state generally exhibited moderate step lengths, no directional persistence, lower proportions of dry time, higher proportions of diving >4 m below the surface, and considerably higher benthic dive counts (posterior median for ?B = 41.4; 95% highest posterior density interval: 41.2–41.7). The resting (?I = ?S = ?L = 9.7; 95% HPDI: 9.4–10.0) and transit (?T = 15.0; 95% HPDI: 14.5–15.5) states exhibited moderate benthic dive counts, while the mid-water foraging state generally had the lowest benthic dive counts (?M = 1.5; 95% HPDI: 1.3–1.7; Fig. 3). The moderate benthic dive counts for the resting and transit states could be attributable to mid-interval state switches, but there certainly could have been opportunistic benthic foraging during transit. There also could have been high-speed, directionally persistent benthic dives while transiting in relatively shallow shelf waters.