Applications of Data-driven Models for Daily Discharge Estimation Based on Different Input Combinations

Accurate and reliable discharge estimation is considered vital in managing water resources, agriculture, industry, and flood management on the basin scale. In this study, five data-driven tree-based algorithms: M5-Pruned model-M5P (Model-1), Random Forest-RF (Model-2), Random Tree-RT (Model-3), Reduced Error Pruning Tree-REP Tree (Model-4), and Decision Stump-DS (Model-5) have been examined to measure the daily discharge of Govindpur site at Burhabalang river, India. The proposed models will be calibrated by daily 10-years time-series hydrological data (i.e., river stage (h) and daily discharge (Q)) measured from 2004 to 2013. In these models, 70% and 30% of the dataset were used for the training and testing stage for the reliability of the developed models. The precision of the models was optimized by investigating five different scenarios based on various time-lags combinations. Model’s performance has been assessed and evaluated using five statistical metrics, namely, correlation coefficient (R²), Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Relative Absolute Error (RAE), and Root Relative Squared Error (RRSE). Results showed that Model-3 outperforms as compared to other proposed models. Machine learning models have been examined five scenarios of input variables during training and testing phases. In comparison of the Model-5 struggled in capturing the river's flow rate and showed poor performance in scenarios where R² metric values ranged from 0.64 to 0.94. Therefore, it can be concluded that the RT model could be used as a robust model for sustainable flood plain management.