Prediction of embankments dam break peak outflow: a comparison between empirical equations and ensemble-based machine learning algorithms

To accurately predict dam break peak outflow (Qp), two standalone [random tree (RT), instance-based k-nearest neighbors learning (IBK)] and four new ensemble machine learning algorithms, that couple the standalone models with two ensemble methods [bootstrap aggregating (BA), disjoint aggregating (DA)], are proposed. The machine learning methods (RT, IBK, BA-RT, BA-IBK, DA-RT, and DA-IBK) and several popular empirical equations are applied to predict Qp using dam reservoir volume above the breach invert (Vw) and height of water in the dam reservoir above the breach invert (Hw) collected from 122 historical dam break events. Three different input scenarios (Vw, Hw, Vw and Hw) are considered to find the most effective combination of input variables. The proposed models are evaluated visually (using scatter and violin plots as well as Taylor diagrams) and quantitatively (using Nash–Sutcliffe Efficiency (NSE), Willmott’s index of agreement (WI), and Legates and McCabe coefficient of efficiency (LM) metrics). It is found that DA-IBK provides the best performance (NSE = 0.866, WI = 0.960 and LM = 0.687), leading to a ~ 28% improvement in NSE compared to the best empirical equation. However, all machine learning models (particularly, the ensemble models), provide substantially better performance than the empirical equations, especially at the highest outflows.