Integration of artificial intelligence to a conceptual hydrologic model

In the presented study, a daily rainfall-runoff model with fewer parameters was developed through the integration of artificial neural network and genetic algorithms available from various artificial intelligence techniques onto a conceptual hydrologic model. The integration of artificial intelligence within the context of the study was achieved on the daily rainfall-runoff model, GR4J (Génie Rural à 4 paramètres Journalier) that is known to be a deterministic, lumped and continuous parametric model. GR4J is a daily rainfall-runoff model which consists of storage and routing elements in its structure while having 4 model parameters as X1, X2, X3 and X4.The study started with the integration of Artificial Neural Networks (ANN) onto the parametric rainfall-runoff model. With the help of this integration, the parameter number of GR4J model was decreased from 4 to 1 and the nonlinear flow routing scheme in the model was modelled with artificial neural networks. Through such integration, a higher model performance was achieved while substantially limiting the number of model parameters.Besides, the automatic calibration of the model parameters by means of genetic algorithms (GA) was performed in the study by integrating these GAs onto the GR4J-ANN integrated model. The resulting GR4J-ANN-GA integrated model including quite few model parameters and having the capability of automatic calibration for its parameters was implemented in the Murat, Selendi, Deliiniş, Demirci, Gördes, Medar and Yiğitler subbasins of the Gediz river basin for exploring the modelling performances. The results indicated that there is considerable increase in the modelling performance as an outcome of the ANN and GA integration with the GR4J model.In conclusion, the study figured out that it is possible to potentially facilitate the calibration by decreasing the number of model parameters through the integration of artificial intelligence techniques such as ANN and GA with conceptual hydrologic models, while indicating that estimation performances can be increased with the use of integrated models as a whole.