Abstract:
Rainfall-runoff modeling is an important tool for water resources management in watersheds and hydrological hazard predictions such as floods. Several research has been carried out by hydrologists to produce efficient models that generate the watersheds’ responses to precipitation. Generally, these models involve parameters that are often unavailable, and even difficult to measure. Therefore, it may be practical to focus on new Deep Learning methods, which are powerful tools that can understand the complexity of the non-linearity relationship between inputs and outputs without having to resort to several parameters.
In this study, the authors used two different models RNN and LSTM on daily data from 5 catchments with a Mediterranean climate where the LSTM model showed better results for what was evaluated by the NSE. Other assessments were made on the LSTM model by RSR and PBIAS where the precipitation and antecedent flow being the parameters that most influenced the model.
