Key points are not available for this paper at this time.
ABSTRACT Accurate daily runoff prediction is essential for water resource management and flood control. However, most models do not consider the nonlinear relationship between meteorological factors and streamflow, which may be a prediction error, especially during high runoff. To solve this problem, this study develops a machine learning-based runoff forecasting framework that incorporates multiple meteorological parameters beyond precipitation. Six ensemble learning models were implemented using daily hydrometeorological data (1960–2022) from the Xin'anjiang River Basin in eastern China. The results show that (1) the incorporation of meteorological elements significantly improves prediction accuracy across all models, with the highest correlation of coefficients (0.955) and Nash–Sutcliffe efficiency (0.905). (2) For flood flow, the gradient boosting model with short-term meteorological inputs showed the highest accuracy. (3) For base and middle flow prediction, the CatBoost model using historical meteorological elements of the past month as input showed the best performance. This study provides evidence that climate conditions beyond precipitation exert influence on runoff generation, particularly during flood events, and offers a new perspective for improving runoff forecasting.
Zhu et al. (Mon,) studied this question.