Fish memories could help predict future occurrence patternsSubmitted by editor on 29 August 2017. Get the paper!
By Jed Macdonald, Kai Logemann, Elias Krainski, Þorsteinn Sigurðsson, Colin Beale, Geir Huse, Solfrid Hjøllo, Guðrún Marteinsdóttir
The phenomena of schooling, shoaling, flocking, and swarming has intrigued ecologists for centuries, inspiring a vast literature devoted to how individual behaviours scale to group-level decisions. For animal groups, choices about when to move, where to feed and breed, or how best to avoid predators are often made collectively, and it is increasingly recognized that such collective decisions can be shaped (at least in part) by social interactions (Keith and Bull 2017), information exchange among individuals and past experiences.
An example of sociality in Atlantic herring.
Within fish schools, neighbouring individuals are usually not closely related, and hence self-interest might act to favour group-level movement decisions in which the majority opinion is adopted. Indeed, both experimental work and simulations have suggested that ‘quorum’ responses, in which the probability of an individual choosing a certain route increases abruptly beyond a threshold number of neighbours choosing that same route, increase decision accuracy, and that larger fish schools, in general, make better, faster decisions; sensu ‘wisdom of the crowd’. These patterns appear to emerge across a wide range of taxa, and for fish, can manifest in improved navigation and capacity to sense dynamic environmental gradients, among other benefits.
Consensus may also be achieved through leadership by a minority of more experienced individuals, or those with strongly held preferences. In some instances, only a knowledgeable few are needed to produce highly accurate movement decisions; however, a complete absence of such leaders may result in poor navigational accuracy. These observations, in conjunction with growing appreciation of the cognitive abilities of group-living fishes, give credence to theories purporting the existence of spatial memory and tradition formation in some species, in which information on previously-used migration routes is thought to be passed down from older, experienced fish to younger, naïve ones, communicated within cohorts and remembered.
To explore these issues, we focused on Atlantic herring (Clupea harengus), a long-lived, dense-schooling species of high commercial importance, noted for its unpredictable shifts in winter distribution. We were particularly interested in the role of collective memory in shaping winter distribution patterns, and developed a series of Bayesian space-time occurrence models, using 23 years of point-referenced fishery and survey data from Icelandic waters, to disentangle its influence from local-scale oceanographic factors, prey availability during the pre-wintering feeding period, the magnitude of recent fishing effort and density-dependence.
Chasing herring aboard Dröfn RE-35 in Ísafjarðardjúp, northwest Iceland.
We found that the previous winter’s occurrence pattern was a strong predictor of the present pattern, its influence increasing with the size of the herring population. Although the mechanisms underpinning this result are uncertain, we suggest that a ‘wisdom of the crowd’ dynamic may be at play, by which navigational accuracy towards traditional wintering sites improves in larger, denser, better synchronized schools. Wintering herring also preferred warmer, fresher, moderately stratified waters of lower velocity, close to hotspots of summer zooplankton biomass, our results indicative of heightened environmental sensitivity in younger cohorts.
The ability of our models to accurately predict the winter occurrence pattern, one-year ahead, highlights the potential of uniting demographic information and dynamic models to quantify both the strength of collective memory in animal groups and its relevance for the spatial management of populations.
Spatial predictions of occurrence probability for four representative winters. For each year, (a) is the mean occurrence probability (ψ) and (b) the sd of ψ (expressed as log-odds) for each grid cell. (c) is the mean intensity of the temporally-independent realization of the spatial random field (w), and (d) is the sd of w. Observed occurrences (black circles) and absences (grey crosses) for each year are overlaid in (a).