Significant variations in event-scale runoff-sediment response mechanisms across Loess Plateau watersheds introduce substantial forecasting uncertainties. This study developed the MESH framework-encompassing flood event identification, multi-dimensional characteristic extraction, and the integration of Spearman correlation, Random Forest, sediment rating curves (SRCs), and hysteresis analysis-to investigate 104 flood events in three typical watersheds on the Chinese Loess Plateau: Baijiachuan-Zhaoshiyao (BJC-ZSY), Wuqi (WQ), and Hongde (HD). Results show BJC-ZSY follows a "high-runoff, high-sediment" regime driven by rainfall, with sediment transport strongly correlated with peak discharge (r = 0.80). Conversely, WQ and HD exhibit "low-runoff, highsediment" characteristics with low runoff efficiency but extreme sediment concentrations averaging 279.26 kg/ m3 and 591.87 kg/m3, respectively. In these sediment-abundant areas, the dominant control on sediment yield modulus shifts from peak discharge to runoff depth. SRC analysis reveals that BJC-ZSY features a low coefficient and high exponent (a = 2.52, b = 0.68), suggesting limited supply but high transport efficiency, whereas HD shows a high coefficient and low exponent (a = 161.43, b = 0.33), indicating abundant supply but lower efficiency. Hysteresis analysis identifies a mixed-source transport model in HD driven by multiple erosion processes. In contrast, counter-clockwise loops dominate in BJC-ZSY (40.0%) and WQ (44.8%), suggesting a prevalence of distal sediment sources. These findings elucidate divergent runoff-sediment relationships and internal mechanisms across the Loess Plateau, providing a scientific basis for targeted regional soil and water conservation management and a robust analytical reference for other globally significant semi-arid regions facing severe erosion.