Daily streamflow modelling using autoregressive moving average and artificial neural networks models: case study of Çoruh basin, Turkey I ˙ brahim Can 1 , Fatih Tosunog ˘ lu 1 & Ercan Kahya 2 1 Civil Engineering Department, Atatürk University, Erzurum, Turkey and 2 Hydraulic Division, Civil Engineering Department, Istanbul Technical University, Maslak/Istanbul, Turkey Keywords artificial neural networks; autoregressive moving average model; Çoruh basin; streamflow. Correspondence I ˙ brahim Can, Atatürk University, Civil Engineering Department, 25240, Erzurum, Turkey. Email: ibcan@atauni.edu.tr doi:10.1111/j.1747-6593.2012.00337.x Abstract Streamflow modelling is a quite important issue for water resources system plan- ning and management projects, such as dam construction, reservoir operation and flood control. This study demonstrates the application of artificial neural networks (ANN) and autoregressive moving average (ARMA) models for modelling daily streamflow in Çoruh basin, Turkey, where there are numerous highly critical power plants either under construction or being projected. Daily streamflow records from nine gauging stations located in the basin were used in this study. In the first phase of our study, ANN and ARMA models were obtained using daily streamflow. In the second phase, 100 synthetic streamflow series were generated using previously determined ANN and ARMA models in order to ensure the preservation of main statistical characteristics of the historical time series. The results have showed that the historical time series have similar statistical parameters to those of the gener- ated time series at 95% confidence level. Introduction Stochastic simulation of hydrological time series has been widely used for solving various problems associated with the planning and management of water resource systems for several decades. Typical examples are the determination of a reservoir capacity, evaluations of adequacy and reliability of a reservoir for a given capacity, evaluation of adequacy of a water resource management strategy under various poten- tial hydrological scenarios, and evaluation of the perform- ance of an irrigation system under uncertain irrigation water distributions (Kim et al. 2004). Therefore, mathematical models are typically needed for stochastic simulation of hydrological processes such as streamflow so that a number of such models have been extensively developed in the last half century. Autoregressive moving average (ARMA) and artificial neural networks (ANN) models have been widely used for streamflow modelling. Both techniques have been success- fully applied to hydrological problems. For example, Zealand et al. (1999) sought the utility of ANN for short-term stream- flow forecasting in the Winnipeg River system in north- western Ontario, Canada. The authors concluded that the ANN significantly outperformed the stochastic-deterministic watershed model. Abrahart & See (2000) compared the fore- casting power of ANN and ARMA models using a 3-year period of continuous river flow data for two contrasting catchments: the Upper River Wye and the River Ouse in York- shire, UK. The authors state that the ANN and ARMA solu- tions provided similar results. Dibike & Solomatine (2001) investigated the applicability of ANN for downstream flow forecasting in the Apure River basin in Venezuela. They imple- mented two types of ANN architectures, namely multilayer perception network and a radial basis function network and compared the performances of these networks with a con- ceptual rainfall run-off model. Ahmed & Sarma (2007) devel- oped ANN models to generate synthetic streamflow series of the Pagladia River, a major north bank tributary of the river Brahmaputra, India. Their results indicated that ANN-based models performed quite successfully in generating synthetic streamflow series. Likewise, there have been a number of studies concerning the ANN and autoregressive models being applied to differ- ent hydrological variables in Turkey during the last decade. Among those recent works, Yürekli et al. (2005) used autore- gressive integrated moving average (ARIMA) models to simu- Water and Environment Journal. Print ISSN 1747-6585 567 Water and Environment Journal 26 (2012) 567–576 © 2012 CIWEM.