Experimental and Artificial Neural Network (ANN) modeling of instream vegetation hydrodynamic resistance

Abstract The present study investigates the effect of upstream structure on the bulk drag coefficient of vegetation experimentally by placing an embankment model with or without moat/depression upstream of the vegetation. The results indicate that in the presence of the upstream structure, the bulk drag coefficient of vegetation is decreased because the upstream structure shares the drag with vegetation. Further, it is noticed that by placing only the embankment on the upstream side, the maximum decrease in the bulk drag coefficient is 11%, and by placing both embankment and moat models on the upstream side of the vegetation, a 20% decrease in the bulk drag coefficient is observed. Based on the variables affecting the bulk drag coefficient, regression models and Artificial Neural Network (ANN) models are developed to predict the bulk drag coefficient. The results from five ANN models with different training functions are compared to find the best possible training function. The coefficient of determination (R2), Root Mean Square Error (RMSE), Nash-Sutcliffe Efficiency (NSE), Sum of Square Error (SSE), Mean absolute error (MAE), and Taylor's diagrams are used to evaluate the performance of ANN modeling techniques. The ANN model having nine neurons in each hidden layer, performs best among the five models, as this model shows the optimal values for the performance indicators, such as the highest R2 and NSE, and minimum values for the RMSE, SSE, and MAE. Finally, the comparison between the regression model and the ANN model shows that the best ANN model, achieving R2 values of 0.99 and 0.98 for the training and validation subsets, outperforms the regression models.