Standalone and Novel Hybrid ML Models for Estimating Stream Velocities in a River Channel with Vegetation

Vegetation present in the flowing water bodies offers a substantial amount of obstruction to the flow. By influencing the flow velocities, these vegetative elements may alter the hydrodynamic characteristics of a river channel. Several laborites as well as field studies have been carried out for a better estimation of vegetation influenced stream velocities. This study investigates the applicability of artificial intelligence-based approaches for the estimation of velocities in a river channel, where the presence of vegetative elements influences a stream’s flow characteristics. To estimate the flow velocities, we made use of a Reduced Error Pruning Tree (REPT), Random Forest (RF), and Random Tree (RT) models as standalone models and used them with hybrid ML models like Multi Scheme (MS) and Random Committee (RC) models. In all, we explored nine ML models, viz., REPT, MS-REPT, RC-REPT, RF, MS-RF, RC-RF, RT, MS-RT, and RC-RT. Performance of the applied approaches has been evaluated by using several performance metrics (Pearson Correlation Coefficient ( R ), Nash-Sutcliffe Efficiency ( NSE ), Agreement Index ( AI ), Root Mean Squared Error ( RMSE ), Scatter Index ( SI ), Percent bias ( PBias ), and Kling-Gupta efficiency ( KGE )). Hybrid ML models outperform their standalone variants as well as the empirical equations. Out of these models, MS-RF ( R=0.957, NSE=0.915, AI=0.977, SI =0.240, PBias=1.597, RMSE=0.072, and KGE=0.928 ) outperformed other approaches in terms of accuracy, followed by RC-RF, RF, RC-RT, MS-RT, RT, RC-REPT, REPT, and MS-REPT.