Predicting Daily Streamflow in a Cold Climate Using a Novel Data Mining Technique: Radial M5 Model Tree

Abstract

In this study, the viability of radial M5 model tree (RM5Tree) is investigated in prediction and estimation of daily streamflow in a cold climate. The RM5Tree model is compared with the M5 model tree (M5Tree), artificial neural networks (ANN), radial basis function neural networks (RBFNN), and multivariate adaptive regression spline (MARS) using data of two stations from Sweden. The accuracy of the methods is assessed based on root mean square errors (RMSE), mean absolute errors (MAE), mean absolute percentage errors (MAPE), and Nash Sutcliffe Efficiency (NSE) and the methods are graphically compared using time variation and scatter graphs. The benchmark results show that the RM5Tree offers better accuracy in predicting daily streamflow compared to other four models by respectively improving the accuracy of M5Tree with respect to RMSE, MAE, MAPE, and NSE by 26.5, 17.9, 5.9, and 10.9%. The RM5Tree also acts better than the M5Tree, ANN, RBFNN, and MARS in estimating streamflow of downstream station using only upstream data.