An improved pattern-guided evolution approach for the development of adaptive individual-based ecological models
Additional Document Info
Ecological models that model species’ adaptation to changing environments may become increasingly important tools for ecologists and environmental scientists faced with the challenges of our changing world. Individual-based models (IBMs) facilitate the modeling of individual diversity and adaptive behaviors. When organisms are modeled with structures that provide inheritable parametric diversity, intergenerational adaptation may also be simulated. These adaptive IBMs may be difficult to calibrate so as to be consistent with field data patterns. The pattern-oriented modeling (POM) calibration approach, whereby model outputs are compared to field data patterns at the end of each simulation, may be limited and computationally expensive under many circumstances. This research further explores an approach, denoted pattern-guided evolution (PGE), that uses field data patterns obtained from published research, to guide the evolution of model organisms within each model simulation. Our preliminary research showed that when demonstrated with an adaptive IBM of an old-field ecosystem, the approach yielded populations of virtual organisms with inheritable parametric diversity, which if well calibrated could potentially be used in future models for simulating adaptive change. However, the model produced in the preliminary studies only partially matched field data patterns, and thus did not confirm the utility of the PGE approach for model calibration. This paper presents three main contributions. Firstly, the paper describes several important improvements to the original approach, which resulted in a model that matched the expected patterns well. Secondly, additional testing was performed to analyze the reusability of the model entities yielded by the approach. Combined, these two contributions confirm the utility of the PGE method for calibrating IBMs for simulating adaptive change. Finally, we estimate that the PGE approach is likely to be ten or more times less computationally costly than that of the conventional POM approach to IBM calibration.