State-specific selection coefficient estimates along with their standard-error (SE) and associated P-value (P) of the HMM-SSF used to predict animal movements according to encamped or travelling mode, in three ecological systems.

A. Habitat selection during the different phases of the migratory behavior, B. and C. Habitat selection during the different phases of the foraging behavior. Estimates of model parameters from a 2-step approach are also shown for the bison dataset (B). Analysis were performed using 13,444, 94,686 (50592 for the 2-step approach) and 72,730 geolocations from 15 mule deer, 42 plains bison and 18 zebras over 2 (2011–2012), 11 (2005–2016) and 3 (2012–2014) years of monitoring, respectively. The predictive capabilities of the HMM-SSF and the 2-step approach (B) were assessed using k-fold validation repeated 100 times, means and ranges of Spearman rank correlation () are reported for each behavioral mode of movement and both observed and random steps. Average travelled distances were calculated using step length and log-transformed step length coefficient estimates (see S1 Appendix).A. 迁徙行为不同阶段的生境选择；B. 与C. 觅食行为不同阶段的生境选择。针对野牛数据集（B），本研究同时给出了基于两步法（2-step approach）得到的模型参数估计结果。本研究分别纳入15头骡鹿、42头平原野牛及18头斑马作为研究对象，三者对应的地理定位数据量依次为13444、94686（两步法分析使用其中50592条）及72730条，监测时长分别为2年（2011–2012年）、11年（2005–2016年）及3年（2012–2014年）。本研究采用重复100次的k折交叉验证，评估了隐马尔可夫模型-步选择函数（Hidden Markov Model-Step Selection Function, HMM-SSF）及两步法的预测性能；针对每种运动行为模式，同时报告了观测步长与随机步长下的斯皮尔曼等级相关系数的均值与范围。平均移动距离通过步长及对数转换步长的系数估计值计算得到（详见附录S1）。