Abstract
The association of conceptual hydrological models and artificial neural networks (ANNs) characterizes a hybrid conformation that represents simultaneously conceptual and non-linear processes related to water flow. This study aimed to evaluate the use of ANNs combined with the conceptual hydrological models IPH II and SAC-SMA to obtain a hybrid model for the estimation of watercourse flows in the Piracicaba River basin, located in the state of Minas Gerais - Brazil. The water flow rates estimated by IPH II and SAC-SMA models were used as input data for the ANN, and a multi-layered perceptron was employed as a neural paradigm. Data from pluviometric, fluviometric, and meteorological stations located in the studied basin and surrounding areas were used for the ANN training and validation. To verify the performance of the hybrid models in estimating water flow, the estimated flow rates were compared with those measured in the fluviometric stations using mean absolute error (MAE), root mean square error (RMSE), bias, Willmott’s concordance index, and Nash-Sutcliffe efficiency index (Ef). The results showed that the use of IPH II model as input data for an ANN increased the accuracy of flow prediction as the hybrid model error decreased. Conversely, if compared to the results from its isolated application, when associated with the ANN, SAC-SMA-model did not improve water flow estimates. Lastly, the hybrid conformation using the IPH II hydrological model improved daily flow estimates of the Piracicaba River basin; therefore, the quality of estimates in a hybrid model depends on the conceptual model used. In addition, the hybrid models had better performance when estimating water flows in larger drainage areas within the Piracicaba River basin.
Keywords:
artificial intelligence; empirical model; accuracy; hydrological modeling
