Review and comparison of performance indices for automatic model induction

Abstract

Abstract One of the more perplexing challenges for the hydrologic research community is the need for development of coupled systems involving integration of hydrologic, atmospheric and socio-economic relationships. Given the demand for integrated modelling and availability of enormous data with varying degrees of (un)certainty, there exists growing popularity of data-driven, unified theory catchment scale hydrological modelling frameworks. Recent research focuses on representation of distinct hydrological processes using mathematical model components that vary in a controlled manner, thereby deriving relationships between alternative conceptual model constructs and catchments’ behaviour. With increasing computational power, an evolutionary approach to auto-configuration of conceptual hydrological models is gaining importance. Its successful implementation depends on the choice of evolutionary algorithm, inventory of model components, numerical implementation, rules of operation and fitness functions. In this study, genetic programming is used as an example of evolutionary algorithm that employs modelling decisions inspired by the Superflex framework to automatically induce optimal model configurations for the given catchment dataset. The main objective of this paper is to identify the effects of entropy, hydrological and statistical measures as optimization objectives on the performance of the proposed approach based on two synthetic case studies of varying complexity.