Uncertainty analysis of an integrated hydrological model using posterior covariance matrix from automatic calibration

This study quantifies the output uncertainties of an integrated hydrological model in the Yanqi Basin, Xinjiang, China. Two cases of model calibration were carried out: Case 1 used four parameters for calibration; Case 2: added fifteen more parameters for calibration. In both cases, hydraulic heads and evaporation patterns from NOAA images were used as conditioning information. In Case 1. no pair of adjustable parameters is highly correlated (highest R: 0.48). In Case 2, several pairs of adjustable parameters are highly correlated (highest R: 0.79). To analyse uncertainty, 5000 parameter realizations were generated using the optimized values and posterior covariance matrix from the model calibration cases. The impact of parameter uncertainty on seven selected model predictions is evaluated. The results show that the uncertainty is underestimated in Case 1. Therefore, it is advisable to include as many as possible adjustable parameters in the model calibration for parameter uncertainty analysis. In this way, it allows scanning of a much larger space of feasible solutions, although generally the calibration tries to use the "parsimony" principle.