Recognition of patterns of optimal diel vertical migration of zooplankton using neural networks

Abstract

Here we apply artificial neural networks to facilitate recognition of regular diel vertical migrations (DVM) of zooplankton in the ocean and lakes, the phenomenon which is considered to be the most significant synchronous biomass movement on Earth. The underlying mathematical framework of finding the optimal (i.e. evolutionary stable) strategies of DVM of zooplankton is based on the generic idea of maximization of fitness of many competing subpopulations each of which uses a particular strategy. To be able to recognize patterns of DVM from data, we have created novel software which technically consists of two interconnected complexes. The first complex is required to find the evolutionarily stable behaviour using the principle of optimality and its implementation produces training samples for the neural network. The second complex provides recognition of evolutionarily stable DVM taking into account a few key characteristics of the aquatic environment and this also allows for some uncertainty (only partial information available) in the input data. In our work, we use a four-layer neural network. Extensive testing of our method demonstrates its efficiency in revealing the presence of detectable regular DVMs as opposed to a random vertical movement of zooplankton.