Abstract
Reliance on basin models by hydrologists to study the effects of changing landscapes on sustainability of water resources is increasing. The success of basin models to provide reliable predictions depends on adequacy of the calibrated model and residual model uncertainty. In this paper, reductions in predictive streamflow uncertainty are demonstrated using streamflow quantities integrated over different sample frequencies as calibration constraints and iterative updating of prior information. Respective model‐quality results in terms of coefficient of model‐fit efficiencies for cases using published parameter values and no calibration, traditional calibration constrained by daily streamflow measurements, and calibration constrained by various derived streamflow quantities were ‐0.44 (very poor), 0.66 (fair), and 0.66 (fair) to 0.91 (excellent). Incrementally reducing the number of active excellent‐quality‐model parameters (from 130 to 6 parameters) reduced the range of uncertainty (from 287.7 m3 s−1 to 54.5 m3 s−1) when predicting the 500‐year peakflow discharge within the Blackberry Creek basin near Chicago, Illinois. The reduction in range of predictive uncertainty, however, coincided with increased model bias for which the 6‐parameter model could not predict the 500‐yr peakflow discharge. Thus, the hydrologist needs to strike a balance between an acceptable level of model complexity that gives rise to predictive uncertainty and model bias.