Automatic calibration of conceptual rainfall‐runoff models: The question of parameter observability and uniqueness

Reasons for the inability to obtain unique and conceptually realistic parameter sets for conceptual rainfall‐runoff models are examined in this paper. The problem is first posed in a framework that allows a more consistent and logical analysis of the related aspects. Response surface studies demonstrate that choice of an objective function that better explains some of the stochastic properties of the errors in the output results in a smoother, better shaped response surface; hence the chances of obtaining unique and realistic parameter estimates are improved. However, our analysis indicates that part of the problem is also due to inadequacies in the model structure. The arguments are illustrated using results obtained with the soil moisture accounting model of the National Weather Service's river forecast system.