Leveraging level data for accurate downstream flow prediction in the Narmada River Basin with advanced machine learning models

ABSTRACT

Accurate prediction of river flow is crucial for water resource management and flood forecasting. This study employs advanced machine learning (ML) models, including CatBoost, Light Gradient-Boosting Machine, Random Forest, and XGBoost, to predict flow at ‘Mandleshwar’ using level data from key locations along the Narmada River Basin. The dataset covers the period from 1978 to 2020 and was sourced from India's Water Resources Information System. Evaluation metrics such as mean squared error (MSE), mean absolute error, root mean square error, normalized root MSE, root mean square percent error, and R-squared (R2) are used to assess model performance. Results show that XGBoost performs well in training, while CatBoost demonstrates superior performance in testing and validation. These findings highlight the potential of advanced ML models in improving flow prediction accuracy, which can inform better water management practices and flood forecasting strategies in the Narmada River Basin.