A Study on Learning-Machine Models for Low-Flow Prediction in River Basins

River discharge can be categorized into low-, drought-, normal-, and high-flow conditions depending on the flow regime and is an essential parameter in water resource management. Traditionally, the quantitative estimation of daily streamflow has primarily relied on hydrological models; however, such models require extensive effort and cost for data preparation and involve complex implementation procedures. To address these limitations, this study developed a predictive framework using hydrometeorological variables to estimate the daily discharge at streamflow monitoring stations. To improve prediction accuracy, particularly in drought-related low-flow conditions, the interquartile range, representing the statistical variability and cumulative distribution characteristics of streamflow data, was applied. In addition, flow regimes were classified into categorical variables to minimize the influence of extreme high-flow values occurring during flood seasons. The proposed methodology was applied to the Weolsan Bridge station in the Geum River Basin and the Seomjingang Dam station in the Seomjingang River Basin. Artificial Neural Network, Random Forest, and Model Tree were employed, and a multiple regression model was applied for comparison. Model performance was evaluated using k-fold cross-validation and assessed based on correlation coefficients, Kendall’s tau, Spearman’s rho, MAE, RMSE, RAE, and RRSE metrics by comparing model predictions with observed values. The results indicated that the models demonstrated higher reproducibility than the regression-based approach. Among them, the Random Forest model achieved the best predictive performance. Overall, the findings suggest that machine-learning models utilizing hydrometeorological variables provide a viable approach for forecasting low-flow conditions and may serve as a useful tool for drought preparedness and water resource management. The results of this study can be used as foundational research data to establish drought response strategies for sustainable water resource management.