COS 139-8
Do interactions matter? A fully-factorial manipulation of four environmental drivers in a microbial aquatic ecosystem

Friday, August 14, 2015: 10:30 AM
321, Baltimore Convention Center
Aurélie M. Garnier, Institute of Evolutionary Biology and Environmental Studies, University of Zurich, Zurich, Switzerland
Owen L. Petchey, Department of Evolutionary Biology and Environmental Studies, University of Zurich, Zurich, Switzerland

In natural ecosystems, multiple environmental drivers occur simultaneously. Limited evidence suggests that interactions between two or three drivers are important (Darling & Côté 2008) affecting all ecological levels of organization (Sala et al. 2000, Crain et al. 2008). But what happens when the number of drivers increases? Using a microbial aquatic ecosystem with algae, bacteria, ciliates and rotifers, I factorially manipulated four environmental drivers with two levels for each: temperature (20°C vs. 25°C), nutrient supply (N:P ratio of 40 vs. 15), organic matter supply (0.23g.L-1 of Protozoan Pellet vs. 0.56g.L-1) and light availability (100% vs. 70% using shade cloth). The experimental design resulted in 16 treatments with six replicates. I observed responses of dissolved oxygen concentration to these factors. From time series, short- and long-term resistance (amplitude of change) and resilience (return time) were estimated. The importance of interactions between treatments was estimated with four-way ANOVA. 


Increases in the number of perturbations, from one to four, were associated with decreased short- and long-term resistance and increased resilience. These associations were significant despite the lack of significance of any interaction terms between treatments in the four-way ANOVA. The number of perturbations explained relatively little variance in the results (between 6 and 41% variance explained), compared to all main effects and interaction terms (between 28 and 93%). Hence the ANOVA results indicate that predictions of average effects of multiple environmental drivers can be based on the additive effects of each driver in isolation. Predicted additive effects were consistently lower than observed effects. This illustrates the danger of applying arbitrary significance thresholds to build predictive models. These insights assist the development of better models for predicting effects of multiple environmental changes on ecological systesms.