The Automatic Calibration of Conceptual Catchment Models Using Derivative‐Based Optimization Algorithms

This paper examines the possibility of using derivative‐based optimization algorithms for automatic calibration of conceptual rainfall‐runoff models with threshold structures. A method is discussed for explicit computation of the derivatives based on an analysis of the modality of behavior present in such models. In the method the threshold structures are not replaced by smoothing functions, thereby preserving the conceptual integrity of these models. The discussion includes a theoretical analysis of a single linear reservoir (with threshold) model which provides some insights into the issue of threshold parameter identifiability. Simulation study comparisons of the Newton (derivative based) and the Simplex (direct search) optimization algorithms indicate that the former is more efficient, especially when the number of parameters to be optimized is large. However, the results indicate that careful study of the issues related to model identifiability is required if any significant progress is to be achieved in the area of automatic calibration of conceptual rainfall‐runoff models.