Comparison of machine learning techniques for rainfall-runoff modeling in Punpun river basin, India

Abstract

Machine learning (ML) models have emerged as potential methods for rainfall-runoff modeling in recent decades. The appeal of ML models for such applications is owing to their competitive performance when compared to alternative approaches, ease of application, and lack of rigorous distributional assumptions, among other attributes. Despite the promising results, most ML models for rainfall-runoff applications have been limited to areas where rainfall is the primary source of runoff. The potential of Random Forest (RF), a popular ML method, for rainfall-runoff prediction in the Punpun river basin, India, is investigated in this paper. The correlation coefficient (R), Root mean squared error (RMSE), Mean absolute error (MAE), and Nash-Sutcliffe efficiency (NSE) are four statistical metrics used to compare RF performance to that of alternative ML models. Model evaluation metrics indicate that RF outperforms all others. In the RF model, we got the best NSE score of 0.795. These findings offer new perspectives on how to apply RF-based rainfall-runoff modeling effectively.