Comparison of Newton‐type and direct search algorithms for calibration of conceptual rainfall‐runoff models

Abstract

An examination of the calibration aspect of conceptual rainfall‐runoff models was undertaken using the Sacramento soil moisture accounting model and a study comparing the performance of a Newton‐type optimization algorithm with that of a direct search algorithm. Results indicate that the direct search algorithm is the more robust of the two because the Newton‐type algorithm is more susceptible to poor conditioning of the response surface. Graphical studies of the response surface of the model's parameter space confirmed the presence of discontinuities and a rough‐textured surface, particularly in the derivatives.