Flow discharge prediction in compound channels by using decision model tree M5

Accurate computation of flood discharge has specific priority for many river engineering projects and flood controls. In flood condition, by flow entering into the floodplains, hydraulic flow is more complex than usual. Recent studies have shown that existing methods do not have sufficient accuracy in calculation of flow discharge in time of flooding. Therefore, many modified methods have been proposed by various researchers. These methods are often time-consuming and sometimes need to use numerical solution of differential equations. In this paper, M5 decision tree model is proposed as a new solution for accurate computation of flow discharge in compound channels. Dimensionless variables of the flow depth ratio, coherence parameter and ratio of computed total flow discharge to bankful discharge have been used as input data and ratio of actual discharge to bankful discharge as output (target) variable. 400 stage-discharge data were collected from 30 experimental and natural compound channels and then divided into calibration and validation groups. Using decision tree model, four linear equations have been derived for calculation of flow discharge in compound channels based on depth ratio. By statistical comparison of results, determination coefficient (R2) and root mean square error (RMSE) were obtained as 0.939 and 0.382 for calibration data and 0.938 and 0.466 for validation data, respectively. The proposed method has more accuracy than the DCM, Ervine et al. (2000) and Haidera and Valentine (2002) methods.