Two integrated conceptual–wavelet-based data-driven model approaches for daily rainfall–runoff modelling

Rainfall–runoff modelling is crucial for enhancing the effectiveness and sustainability of water resources. Conceptual models can have difficulties, such as coping with nonlinearity and needing more data, whereas data-driven models can be deprived of reflecting the physical process of the basin. In this regard, two hybrid model approaches, namely Génie Rural à 4 paramètres Journalier (GR4 J)–wavelet-based data-driven models (i.e., wavelet-based genetic algorithm–artificial neural network (WGANN); GR4 J–WGANN1 and GR4 J–WGANN2), were implemented to improve daily rainfall–runoff modelling. The novel GR4 J–WGANN1 hybrid model includes the outflow (QR) and direct flow (QD) obtained from the GR4 J model, and the GR4 J–WGANN2 hybrid model includes the soil moisture index (SMI) obtained from the GR4 J model as input data. In hybrid models, wavelet analysis and the Boruta algorithm were implemented to decompose input data and select wavelet components. Four gauging stations in the Eastern Black Sea and Kızılırmak basins in Turkey were used to observe modelling performance. The GR4 J model exhibited poor performance for extreme flow forecasting. The novel GR4 J–WGANN1 approach performed better than the GR4 J–WGANN2 model, and the hybrid models improved modelling performance up to 40% compared to the GR4 J model. In this regard, integrated conceptual–wavelet-based data-driven models can be useful for improving the conceptual model performance, especially regarding extreme flow forecasting. HIGHLIGHTS Two hybrid models (i.e., GR4 J–WGANN1 (Génie Rural à 4 paramètres Journalier-wavelet–genetic algorithm–artificial neural network) and GR4 J–WGANN2) and the GR4 J conceptual model were implemented for rainfall–runoff modelling.; The Boruta algorithm indicated a higher importance level of outputs obtained from the GR4 J model as input data than other variables in hybrid models.; The hybrid models improved the performance by up to 40% compared to the GR4 J model.; The proposed GR4 J–WGANN1 model yielded better results than the GR4 J–WGANN2 model.;

Abstract licence: CC BY-ND

A person counts once a day per paper. We keep no cookie and no address: a reader is a hash that changes every day. Any month under five reads shows no number.