High-resolution spatio-temporal quantification of fish predator-prey interactions over ecosystem scales with multispectral underwater sensing and optimality of human visual perception with natural daylight

Abstract

Marine ecosystems face simultaneous pressures from human activities, ocean industrialization, potential global warming and changing habitats. Continuous monitoring of marine biodiversity and ecosystem processes is needed to assess the individual fish species survivability in such conditions. The increasing use of computer modeling and simulations based on significantly under-sampled data of the marine environment, however, leads to unconstrained and potentially unstable predictions of key processes. To address this issue, we demonstrate a technology enabling synoptic quantification and distinction of multispecies fish population densities over ecosystem scales with continuous spatial and temporal resolution. This enables high-resolution quantification of predator-prey interactions in space and time over ecosystem scales. We present an example of an event in the Barents Sea where a massive cod predatory swarm of approximately 1.9 million individuals attacks a defending coherently moving linear capelin prey structure extending over 14 km containing approximately 23 million individuals. Capelin are a keystone species of the Arctic ecosystem. Cod are their primary predator, but cod populations have collapsed everywhere except in the Nordic Seas due to overfishing causing significant changes in ecosystem balance in those regions. We provide high-resolution spatial density images finely sampled over time of cod convergence on capelin prey, estimated capelin consumed, capelin survived and satiated cod predators quantifying the detailed spatio-temporal dynamics of predation. From these we estimate 58% of the entire capelin group was consumed by the swarming cod within 4 hours where the detailed imagery of behavioral shoal structure show capelin in the highest density regions have the highest probabilities of survival. Other interactions we quantified between predatory juvenile cod and pre-spawning capelin groups indicate a variety of behavioral mechanisms with varying levels of efficiency are at work for both the predators and prey over the large scales observed here. These observations are made with multispectral ocean acoustic remote sensing which enables instantaneous imaging of fish populations over thousands of square kilometers with average spatial resolution on the order of 100 m and temporal resolution of about 1 minute. Wide-area species classification and simultaneous population density estimation of individual species employs sensing frequencies at or near fish swimbladder resonance where the large differences across fish species are discernible. Such synoptic imaging at areal rates roughly 10^4 to 10^6 times greater than conventional methods may lead to more stable prediction of key ecosystem processes and has broad applications in remotely classifying fish populations, studying ecosystem functions and assessing species sustainability.

Patterns in light intensity contain vital information for organisms that utilize visual sensory perception for survival in their environments. Psychophysical experiments on visual intensity discrimination with artificial light sources over a century have shown that the smallest detectable change in light intensity, termed just-noticeable difference, grows roughly in direct proportion to the stimulating intensity, approximately following Weber’s law of perception. The potential advantages of Weber’s law in the context of sensing and pattern recognition, however, have not been quantified given the natural intensity scintillation of environmental light. Here we find Weber’s Law to be a consequence of attaining the theoretical minimum mean-square error possible, the Cramer-Rao lower bound, in resolving the intensity of naturally scintillating light. We first obtain the statistics of environmental light signals which we find naturally scintillate with a standard deviation proportional to mean intensity. Given our natural scintillating light intensity data, we find log-transformed intensity and Fechnerian transformed intensity are equivalent to variance-stabilizing-transformed intensity. We then find intensity resolution that follows Weber’s Law is statistically optimal in pattern recognition by simple matched-filter correlation and maximizes information reception by homeomorphically transforming signal-dependent intensity scintillation to signal-independent Gaussian noise which can be canceled without loss of signal information. We show just-noticeable-differences in light intensity obtained from psychophysical experiments with artificial light approximately attain the Cramer-Rao lower bound on intensity resolution expected from our observed natural light intensity scintillation. Human intensity resolution is in this manner approximately optimally adapted to the statistical properties of natural light scintillation with Weber’s Law as a consequence. Along these lines, the same kind of variance-stabilizing transformation is used in the first part of the thesis in acoustic sensing of fish due to intensity scintillation of measured acoustic intensity data converged upon by the central limit theorem.