Hybrid models based on mode decomposition and recurrent neural networks for streamflow forecasting in the Chira river in Peru

Streamflow forecasting at short horizons is vital for the management of water resources. However, the streamflow behaviour is non-linear and not stationary. To address this challenge, artificial intelligence techniques have been used to increase accuracy. Additionally, signal decomposition techniques such as empirical mode decomposition, ensemble empirical mode decomposition, and variational mode decomposition, have been applied in different fields as a pre-processing stage prior to modelling to improve forecasting. This study evaluates the effect of the aforementioned decomposition techniques used with a recurrent neural network called long short-term memory to increase the precision of the daily prediction of the Chira river streamflow in northern Peru, characterized by a special dynamic due to a strong seasonal behavior and the influence of the El Nino–Southern Oscillation (ENSO).