Using Multiagent Systems to Develop Individual-Based Models for Copepods: Consequences of Individual Behavior and Spatial Heterogeneity on the Emerging Properties at the Population Scale

Abstract

In marine ecosystems the production of planktonic copepods supports most food webs, directly affecting higher trophic levels (including pelagic fish populations) and the biological pump of carbon into the deep ocean. 1 Traditionally, in ecological models dealing with pelagic processes, the lowest trophic levels have been represented with functional groups. Except when size is treated explicitly, these models assume that all individuals within a trophic compartment are identical. 2 Hence, the prevailing approach in observing and modeling pelagic populations is to characterize average processes (such as feeding, birth, growth, migration, and mortality) that are acting on total abundances or on spatially distributed fields of concentrations. 3–5 However, several studies have shown that zooplankters, including copepods, possess individual traits and behaviors; 6–8 consequently they act and react individually. Furthermore, the individuals may encounter different patterns of food and physical parameters, which may increase the inter-individual variability. 9 At the population level the individual variability on the molting rate of copepods was observed from experimental work. 10,11 The new emerging individual-based modeling (IBM) techniques are appropriate to cope with such variability. 12 The success of this approach can be attributed to the fact that IBMs make more realistic assumptions than the more traditional state variable models. The IBM is a “bottom-up” approach, and may be opposed to the classical state variable modeling, which is a “top-down” approach. 13,14 Uchma´ nski and Grimm 15 defined individual-based models as models that should include discrete individuals with their complete life cycles as well as differences between individuals, and the dynamics of the resources (in general with spatial heterogeneity) over which the individuals compete. A corollary to this definition, which is a fundamental aspect of IBM, is the explicit representation of individual variability. 16 The differences among individuals affect the dynamics of the population. The interactions between individuals and/or their environment lead to global properties at the population level and/or at large spatial scales. This property referred to as “emergence” represents another fundamental concept of the IBM approach. 17 The last decade (1990s) has been characterized by an exponential increase in the number of publica- tions using IBMs in ecology. 18 Certainly the increase in computer power and the development of object computer languages and associated tools as simulators have significantly contributed to the rise of the IBM. 19 Nevertheless, the creation and execution of an IBM is still a delicate operation that requires substantial computational investment. 20 Therefore, there is often a distance, if not a gap, between modelers and end users, which may have negative consequences. Even if some problems still exist, caused by the complexity of individual-based models, 12 the recent progress in developing specialized platforms and tools has increased the accessibility of this approach to a broader range of scientists who may or may not have previous modeling experience. 14 In this chapter we use a recently developed platform based on multiagent systems and specialized in the development of population dynamics models. 20 This platform, called Mobidyc , is designed to make the end users true masters of their models without the assistance of computer experts. This new platform has been used to introduce an educational example that shows how IBM development can lead to new perspectives on our understanding of aquatic ecosystems structures and functions. First, Mobidyc has been used to build easily and without any hard-coding a model of the whole life cycle of the copepod Centropages abdominalis . We subsequently introduced (1) cannibalism, (2) predation, and (3) explicit spatial patterns to study their consequences on C . abdominalis at the population level. The results obtained here can be extended to other populations and/or organisms. In particular, within this platform the user can focus on model development and running experiments rather than hard-coding the model, which needs qualifications in computer programming that the biologist has not yet attained.