A novel fully hybrid simulation-optimization approach for enhancing the calibration and verification performance of the TUW hydrological model

Abstract

In this study, a novel fully hybrid simulation-optimization approach is proposed for enhancing the prediction performance of the conceptual Technische Universität Wien (TUW) hydrological model. In the simulation part of the proposed approach, the TUW model is hybridized with a support vector regression (SVR) model where SVR directly uses the TUW model outputs as input and simulates the runoff process. This hybridized simulation model is then integrated to a hybrid optimization approach where heuristic harmony search (HS) and Nelder-Mead Simplex (NMS) optimization approaches are mutually integrated. The objective of the hybrid HS-NMS optimization approach is to calibrate the associated parameter values of both TUW and SVR models by maximizing the Nash-Sutcliffe efficiency (NSE) calculated between simulated and observed runoff values. The applicability of the proposed approach is evaluated on a Murat sub-basin of the Gediz River Basin (GRB) in Turkey for different simulation and optimization model combinations. Identified results indicated that the proposed fully hybrid simulation–optimization approach not only efficiently calibrates the associated solution parameters, but also improves the prediction performance of the TUW model.