Ideally, general ecological relationships may be used to predict responses of natural communities to environmental change, but few attempts have been made to determine the reliability of predictions based on descriptive data. Using a previously published structural equation model (SEM) of descriptive data from a coastal marsh landscape, we compared these predictions against observed changes in plant species density resulting from field experiments (manipulations of soil fertility, flooding, salinity, and mammalian herbivory) in two areas within the same marsh.
In general, observed experimental responses were fairly consistent with predictions. The largest discrepancy occurred when sods were transplanted from high- to low-salinity sites and herbivores selectively consumed a particularly palatable plant species in the transplanted sods. Individual plot responses to some treatments were predicted more accurately than others. Individual fertilized plot responses were not consistent with predictions (P > 0.05), nor were fenced plots (herbivore exclosures; R2 = 0.15) compared to unfenced plots (R2 = 0.53). For the remaining treatments, predictions reasonably matched responses (R2 = 0.63).
We constructed an SEM for the experimental data; it explained 60% of the variance in species density and showed that fencing and fertilization led to decreases in species density that were not predicted from treatment effects on community biomass or observed disturbance levels. These treatments may have affected the ratio of live to dead biomass, and competitive exclusion likely decreased species density in fenced and fertilized plots. We conclude that experimental validation is required to determine the predictive value of comparative relationships derived from descriptive data.