Development of new machine learning model for streamflow prediction: case studies in Pakistan

Abstract

For accurate estimation of streamflow of a mountainous river basin, a novel hybrid method is developed in this study, where gradient-based optimization (GBO) algorithm is employed to adjust adaptive neuro-fuzzy system's (ANFIS) hyperparameters. Two key mountainous basins in Pakistan, namely, Gilgit and Astore basins, were selected to show the model's effectiveness in predicting monthly streamflows using temperature and antecedent streamflow data. Several benchmark methods for optimizing ANFIS parameters were compared, which includes particle swarm optimization (PSO), genetic algorithm (GA), differential evolution (DE), ant colony optimization (ACO) and grey wolf optimization (GWO). The GBO algorithm enhanced ANFIS prediction accuracy more than the other benchmark methods. ANFIS-GBO improved the prediction accuracy of other benchmark algorithms hybrid ANFIS models by about 28–29.5, 28–26.9, 28.7−40.4, 11.2–20.1 and 10.4–20.9% with respect to root mean square error (RMSE), normalized RMSE, mean absolute error, determination coefficient and Nash Sutcliffe Efficiency in the studied basins, respectively. The ANFIS-GBO model also improved the peak streamflow prediction accuracy of ANFIS-GWO, ANFIS-PSO, ANFIS-ACO, ANFIS-GA and ANFIS-DE by about 4–23.7, 23.1–32.4, 15.4–41.6, 18.9–49.4 and 17.2–52.3%, respectively. The ANFIS-GBO also showed higher strength than the other models in estimating streamflows from nearby station data as input. Performance comparison of GBO based ANFIS hybridized model with standalone ANFIS model showed that GBO successfully enhanced the prediction accuracy of ANFIS model by optimal tuning of its parameters.