Contextual interactions can increase unexplained random variability in ecosystem-level studies

Contextual interactions arise from the nonadditive relationships between model factors at the ecosystem level and are represented as statistical interactions in multi-factor experimental designs. Because ecosystem experiments are rarely fully replicated, little is known about how important a role contextual interactions play in ecosystem response. Thus ecosystem-level models may contain substantial bias that is interpreted as random error. The component of variability in ecosystem response due to contextual interactions was quantified using calcium, magnesium, potassium and volume of leachate export as dependent variables in a factorial mesocosm experiment located at two locations in Vermont. The mesocosm treatments consisted of three factors that influence nutrient cycling in forest ecosystems, site (climate), soil, and tree community. The average annual flux for the mesocosms was 0.81 g m^-2 of K, 2.94 g m^-2 of Mg, and 18.16 g m^-2 of Ca. The variance of total annual nutrient flux has been partitioned between the main effects, contextual interactions, and random effects using an analysis of variance. Results show that the main factors controlled 98 % of variation in Ca export, 55 % of variation in Mg export, 78 % of variation in K export and 64 % of variation in leachate volume. Contextual interactions were found to have significant control over the variation in export of Ca, Mg, K and volume of leachate, controlling 0.4, 38, 8, and 17 % respectively. The percent variation due to random effects was 1 % for Ca, 5 % for Mg, 13 % for K and 18 % for leachate volume. These results support our hypothesis that contextual interactions can control a significant portion of total variability in ecosystem response, which may result in an overestimate of the random variability in ecosystem-level field studies.