Abstract
Animal groups need to be sensitive to important information, but robust to irrelevant signals and changes in group structure. Many species of fish, birds, and ungulates use visual cues to coordinate their motion as a group, search and find food, and spread information about predation threats through the group. How are animal group communication networks structured, and how do they facilitate fast and efficient information processing within a group? Here we use experiments with fish, combined with visual field reconstruction, motion correlation, and network modeling to ask how visual networks influence motion, and how network structure affects information transfer within the group.
Although interactions in animal groups can be approximated by proximity networks, a more accurate representation of inter-individual connections can be obtained via visual field reconstruction \cite{Strandburg_Peshkin_2013,Rosenthal_2015}. This reveals the 'structural' network, i.e. what communication routes may exist between individuals. The 'functional' network, which represents the actual exchange of information, is constrained by structure, but may have different patterns of activity. We track the motion of small schooling fish (three-spined stickleback, Gasterosteus aculeatus) to map structural-functional relationships and ask how visual cues are used in coordinating collective motion. Bar-coded tracking enables individuals to be reliably identified during group motion (Fig 1A). Raycasting (Fig 1B) is used to construct visual networks. Pairwise velocity correlations (Fig 1C) are used to form a functional network for how motion decisions spread in the group \cite{Nagy_2010}. We ask if a leadership hierarchy exists during typical schooling motion, and if community detection algorithms can predict group fission events. We features of compare the time-dependent fish visual and motion correlation networks to random, small-world, and other network models, and use simulations to ask how the measured network characteristics of clustering, persistence/switching times of connections, and degree distributions affect information spread, network robustness, and collective decision-making processes.