Abstract: The scarcity of sufficient streamflow observations,and in some cases,ground-based precipitation measurements in small watersheds poses substantial challenges for reconstructing long-term streamflow series and estimating design floods.In recent years,the increasing availability of global precipitation products derived from satellite observations and atmospheric reanalysis has provided new opportunities for hydrological applications in data-scarce regions.This study investigated an ungauged small watershed—the Yangyi Reservoir watershed in Wenzhou,Zhejiang Province,China,and developed a local correction framework for the Multi-Source Weighted-Ensemble Precipitation (MSWEP) dataset.The corrected MSWEP precipitation was then applied to estimate long-term streamflow series and to calculate design floods in the ungauged small watershed.The results demonstrate that the Feature-Enhanced Random Forest (FERF) algorithm effectively improves the consistency between MSWEP and gauge-based precipitation at both daily and seasonal scales,leading to enhanced accuracy in local precipitation estimation.Compared with the traditional neighboring-gauge transplantation methods,the corrected MSWEP dataset shows some potential in streamflow estimation and design flood calculation.For return periods ranging from 1000 to 50 years,the design peak discharges derived from corrected MSWEP precipitation are 8.51%~11.33% lower than those obtained using Quxi Gauge-transplanted precipitation,while the corresponding 3-day flood volumes are underestimated by 4.95%~7.22%.