Analysis of uncertainty in a salt marsh carbon cycling model: A comparison of methods

Abstract

Both linear sensitivity analysis and the comparison principle are used to analyze the effect of parameter bias (constant error in the parameters) on the predictions of the Spartina subcomponent of a salt marsh carbon cycling model. A stochastic version of the model and a linear approximation to it are both used to analyze the effects of variance (randomly fluctuating error). Bias is the more serious problem, and the linear sensitivity analysis gives poor error estimates for 10% parameter bias compared to the exact results from the comparison principle, although it does well for 1% parameter bias. Error from bias can be reduced by including thresholds and carrying capacities directly as model parameters. Linear approximations do a fairly good job of predicting the effects of variance. Error from both sources increases rapidly during periods of rapid Spartina growth, but decreases as the system approaches equilibrium.