PREDICTION IN COMPLEX COMMUNITIES: ANALYSIS OF EMPIRICALLY DERIVED MARKOV MODELS

Abstract

Understanding and predicting the effects of species extinction and other environmental impacts in natural ecosystems is difficult because of their inherent complexity. Successful prediction in complex systems requires empirically accessible theoretical frameworks to limit the range of possible outcomes. I explored the predictions of multispecies Markov models, based on transition probabilities derived from repeated censuses of fixed points in the middle intertidal community of Tatoosh Island, Washington, USA, under natural conditions and after altering the model to reflect species extinction and changes in size-structured processes. The unmanipulated model predicted a mussel-dominated community and quantitatively captured 98.7% of the variation in community composition in independent transects. Although model assumptions that transition probabilities did not vary over time and depended only on the state of a point at the previous census were violated, altering the model to account for these violations provided no improvement in model performance. The manipulated models indicated that most species and transitions had relatively weak effects on community composition, the strongest effects being associated with California mussels (Mytilus californianus), articulated coralline algae (Corallina vancouveriensis), and predatory birds. The manipulated models also predicted that extinction of several species, particularly predatory birds, affects the resilience of the system, as assessed by return time. The effects of species on composition and resilience were poorly correlated, suggesting that assessing the importance of species depends on the type of ecosystem response being considered. The model successfully predicted quantitative shifts in community composition and ecosystem dynamics under several novel conditions: on vertical walls where bird predation and adult mussel recruitment are absent, and in previously published experiments that excluded birds from horizontal benches. These results suggest that Markov models can provide a useful predictive framework for some types of environmental impacts on complex ecosystems.