Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
4576164 | Journal of Hydrology | 2013 | 10 Pages |
•We compared WANN and LGP methods for monthly streamflow prediction.•LGP performs better than WANN in low and medium flows prediction.•Adding more input signals might lead the WANN models to less accurate results.•WANN is more capable of capturing global maximums (annual peaks) than LGP.•LGP is found to be more applicable than WANN for monthly streamflow prediction at Çoruh River.
SummaryAccurate prediction of streamflow is an essential ingredient for both water quantity and quality management. In recent years, artificial intelligence (AI) techniques have been pronounced as a branch of computer science to model wide range of hydrological processes. A number of research works have been still comparing these techniques in order to find more efficient approach in terms of accuracy and applicability. In this study, two AI techniques, including hybrid wavelet-artificial neural network (WANN) and linear genetic programming (LGP) technique have been proposed to forecast monthly streamflow in a particular catchment and then performance of the proposed models were compared with each other in terms of root mean square error (RMSE) and Nash–Sutcliffe efficiency (NSE) measures. In this way, six different monthly streamflow scenarios based on records of two successive gauging stations have been modelled by a common three layer artificial neural network (ANN) method as the primary reference models. Then main time series of input(s) and output records were decomposed into sub-time series components using wavelet transform. In the next step, sub-time series of each model were imposed to ANN to develop WANN models as optimized version of the reference ANN models. The obtained results were compared with those that have been developed by LGP models. Our results showed the higher performance of LGP over WANN in all reference models. An explicit LGP model constructed by only basic arithmetic functions including one month-lagged records of both target and upstream stations revealed the best prediction model for the study catchment.