Streamflow reconstruction at the source of the Yangtze River using ensemble machine learning

Study region The source region of the Yangtze River (SRYR) Study focus This study reconstructed an 813-year annual streamflow series utilizing ensemble machine learning and the Monsoon Asia Drought Atlas Version 2 (MADAv2) dataset. A stacking model integrating Random Forest (RF) and extreme gradient boosting (XGBoost) was developed and demonstrated optimal performance. Based on the reconstructed series, we systematically investigated long-term trends, multiscale periodic characteristics, and underlying driving mechanisms of streamflow in the SRYR. New hydrological insights for the region The reconstruction identified pronounced dry and wet periods as well as extreme hydrological events in the SRYR, which aligned well with documented historical records. Wavelet analysis revealed significant cyclical variations at multiscale cycles (20–375 years). The streamflow variations are influenced by external forcing (e.g., solar activity) and internal climate variability (e.g., PDO and ENSO). This study presents a novel reconstruction framework that enhances our understanding of long-term hydrological variability and its climatic drivers.