Deep Learning Approach with LSTM for Daily Streamflow Prediction in a Semi-Arid Area: A Case Study of Oum Er-Rbia River Basin, Morocco

Daily hydrological modelling is among the most challenging tasks in water resource management, particularly terms of streamflow prediction semi-arid areas. Various methods were applied order to deal with this complex phenomenon, but recently data-driven models have taken a better space, given their ability solve problems time series. In study, we employed Long Short-Term Memory (LSTM) network simulate daily over Ait Ouchene watershed (AIO) Oum Er-Rbia river basin Morocco, based on temporal sequence situ and remotely sensed hydroclimatic data ranging from 2001 2010. The analysis adopted work three-dimension input required by LSTM model (1); samples used three splitting approaches: 70% dataset as training, considering year cross-validation method; (2) length; (3) features using two different scenarios. results demonstrate that performs poorly default scenario, whereas best during testing found length 30 days approach 3 (R2 = 0.58). addition, fed lagged scenario Forward Feature Selection (FFS) method provides high performance accuracy 2 0.84) 20 days. Eventually, applications related resources management where are limited, use deep learning technique able create predictive accuracy, which can be enhanced right combination subset FFS.