A novel smoothing-based long short-term memory framework for short-to medium-range flood forecasting
Amina Khatun, Chandranath Chatterjee, Gaurav Sahu, Bhabagrahi Sahoo
Abstract
A novel smoothing-based long short-term memory (Smooth-LSTM) framework for flood forecasting up to five days ahead is proposed, and compared with the benchmark LSTM (LSTM) model, an Artificial Neural Network (ANN) model, and the conceptual Nedbør Afstrømnings Model (MIKE11 NAM)-Hydrodynamic (HD) (MIKE) hydrological model. This framework was tested in the typical middle Mahanadi River basin (India), which has a tropical monsoon-type climate. Variation of training loss indicated the LSTM network has a higher learning ability at smaller network and batch sizes. The Smooth-LSTM model could predict streamflow with higher Nash-Sutcliffe efficiency of 0.82–0.87 at up to five days lead time with a better reproduction of the observed crucial high peak floods, whereas the corresponding MIKE, ANN and LSTM model-based forecasts were acceptable only up to four-, three- and one-day lead times, respectively. Overall, the Smooth-LSTM model is found to be robust in operational flood forecasting, with lower uncertainty and the least sensitivity to redundant input information.