A novel approach for rainfall-runoff modelling using a biogeography-based optimization technique: Fid-Bau Portal

A novel approach for rainfall-runoff modelling using a biogeography-based optimization technique

Roy, Bishwajit / Singh, Maheshwari Prasad / Singh, Anshuman

This study adopted a novel biogeography-based optimization (BBO) using an adaptive neuro-fuzzy inference system (ANFIS) technique that is BBO-ANFIS, for one day ahead runoff forecasting. Further, to check robustness of BBO-ANFIS, a comparative study has been done with two well-known hybrid models genetic algorithm-based ANFIS (GA-ANFIS) and firefly-based ANFIS (FA-ANFIS). For model development, two input features (rainfall and runoff), i.e. historical daily accumulated mean rainfall at 1 km resolution and mean daily discharge of three river catchments (specifically, river Fal at Tragony station, river Seaton at Trebrownbridge station and river Kenwyn at Truro) are considered. For the performance evaluation of models, a range of model performance indicators (correlation coefficient (r), mean absolute error (MAE), Nash–Sutcliffe efficiency coefficient) have been used. From results, it was found that the BBO-ANFIS model (r = 0.93, MAE = 0.29, NSE = 0.86 for Fal at Tragony, r = 0.96, MAE = 0.09, NSE = 0.92 for Seaton at Trebrownbridge and r = 0.93, MAE = 0.05, NSE = 0.87 for Kenwyn at Truro) has the best performance than the GA-ANFIS model (r = 91, MAE = 0.29, NSE = 0.82 for Fal at Tragony, r = 0.91, MAE = 0.10, NSE = 0.91 for Seaton at Trebrownbridge and r = 0.92, MAE = 0.06, NSE = 0.85 for Kenwyn at Truro) and the FA-ANFIS model (r = 0.91, MAE = 0.34, NSE = 0.82 for Fal at Tragony, r = 0.95, MAE = 0.10, NSE = 0.91 for Seaton at Trebrownbridge and r = 0.93, MAE = 0.06, NSE = 0.86 for Kenwyn at Truro) for rainfall-runoff (R-R) modelling, This study shows the sensitivity test of standalone machine learning model parameters using three metaheuristic techniques and identifies the best technique that can be efficiently used in hydrological modelling.