Application of Machine Learning Models for Improving Discharge Prediction in Ungauged Watershed: A Case Study in East DuPage, Illinois

Abstract

Accurate flood prediction models and effective flood preparedness rely on thoroughly understanding rainfall–runoff dynamics. Similarly, effective rainfall–runoff models account for multiple interrelated parameters for robust runoff prediction. Process-based physical models offer valuable insights into hydrological processes, but their effectiveness can be hindered by data limitations or difficulties in acquiring specific data. Motivated by the frequent flooding events and limited data availability in the East Branch DuPage watershed, Illinois, this study addresses a critical gap in research by investigating effective discharge prediction methods. In this study, two significant machine learning (ML) models, artificial neural network (ANN) and support vector machine (SVM), were employed for discharge prediction. Historical data spanning from 2006 to 2021 were utilized to assess the performance of the models. Hyperparameter tuning was performed on the models to optimize their performance, and root mean square error (RMSE), Nash–Sutcliffe efficiency (NSE), percent bias (PBIAS), coefficient of determination (R2), and the normalized root mean squared error (NRMSE) were used as evaluation metrics. Although both machine learning models demonstrated strong performance, the analysis revealed that the ANN model emerged as the more reliable option for predicting discharge in the watershed. Crucially, the ANN model surpassed the SVM model’s performance, achieving superior accuracy in predicting peak discharge events within the study area. Our findings have the potential to assist decision-makers and communities in implementing more dependable flood mitigation strategies, particularly in regions where hydrology data are limited.