Confronting the indirect effects of controlled fire: the impact of vegetation shifts on carbon stocks of oak-hickory forests
Plant functional traits have been increasingly used to predict shifts in ecological processes with environmental change because of their ability to provide mechanistic insights on the biotic controls of biogeochemical cycles. The utility of traits in predicting the impact of changing land management practices, such as prescribed fire, has yet to be well addressed. This gap in our understanding may be due to the methodological difficulties in accounting for interspecific as well as intraspecific trait variation, which together may provide much more accurate assessments of the effects of management on ecosystem function. We were interested in two questions: (1) Does the inclusion of intraspecific trait data improve our ability to predict soil carbon stocks, a key ecosystem service, across forests managed with fire and (2) If we use widely available interspecific trait data instead, will we retain the same predicative ability? We compared the utility of fire-based management status, abiotic characteristics, interspecific and intraspecific functional trait variation in predicting changes in soil carbon stocks in central hardwood forests. Intraspecific trait values were collected in the field from the 85% most abundant understory species, whereas interspecific traits were sourced from the field and the TRY Database.
Particulate organic matter (POM) ranged from 594 to 1872 g m-2, with fire-managed stands averaging significantly higher (15%) POM stocks than unburned stands (β = 262.48, SE ± 168, p = 0.0013). Despite fire management having a strong overriding effect on soil carbon, the full model containing management, abiotic and interspecific functional trait predictors explained significantly more of the variation in particulate organic matter stocks than the management-only and null models, according to AICc (p < 0.0001, K = 10, wi = 0.999, Table 1). All models explained significantly more variation than the null model (p < 0.0001), but the reduced models (containing only abiotic or plant functional trait variables) were overall poor predictors of POM in comparison to the full model (wi < 0.0001, Table 1). Our random intercept term based on the plot in which the data were sampled accounted for 173.46 in residual variance of our response variable, leaving 307.12 in residual variance. The variance structure partitioned between management strata showed that burned stands had on average 3.25 times more residual variance in POM stocks than unburned controls (varIdent (burned) residuals: 305.12, varIdent (unburned) residuals: 93.76).