Meeting Abstract
Ecologists have long sought to understand the dynamics of populations and communities by deriving mathematical theory from first principles. This standard theoretical approach differs radically from the data mining methods making their way into ecology from fields like machine learning. Here, we explore whether a data-mining technique known as symbolic regression can bridge these disparate approaches, by automatically discovering dynamical relationships in real ecological data, but expressing those relationships using dynamical equations – the language of theoretical ecology. By applying this technique to three classic demographic time series we found this method rapidly discovers dynamical models that explain most of the variance in all time series. Model predictive ability begins to saturate with increasing model complexity at a surprisingly small number of free parameters and the model occupying the saturation point was precisely the model previously proposed by theoretical ecologists: the logistic growth equation for Paramecium growing in isolation, the Lotka-Volterra predator-prey equations for Paramecium and Didinium in co-culture, and a chaotic stage-structured population model for Tribolium flour beetles. Our findings suggest a powerful new way to merge ecological data analysis and theory development. Furthermore, symbolic regression may have useful applications in organismal biology, for example by reverse engineering ‘behavioral circuits’ from time series of animal movements and behaviors.
