River runoff and suspended sediments exhibit high dynamic and multiscale variability, particularly during rainstorm-driven floods in arid and semiarid regions. Despite the widespread use of the hysteresis loop model, flood-event-scale hysteresis patterns and their controls remain poorly quantified in these environments. This study investigated the hysteresis patterns between runoff and sediment and identified the dominant factors affecting the hysteresis index (HI) in a typical semiarid catchment in northern China. Forty-eight flood events were classified by K-means clustering using the runoff depth (H), flood duration (T), and peak flood flow (Qp) into three types: A (low-intensity, medium-duration floods), B (medium-variability, long-duration floods), and C (high-intensity, short-duration floods). The monthly runoff and sediment load were decomposed via multivariate empirical mode decomposition (MEMD) to extract scale-specific dynamics, whereas event-scale hysteresis patterns were quantified separately to assess sediment transport mechanisms during floods. Figure-eight loops occurred most frequently (31.3 %), followed by counterclockwise (29.2 %), complex (22.9 %), and clockwise (16.7 %) loops. HI is governed by distinct hydrological factors for each loop type: flood peak timing skewness (FT) for clockwise loops; Qp, suspended sediment yield (SSY), mean suspended sediment concentration (Sm), and peak suspended sediment concentration (Sp) for counterclockwise loops; T and high traffic duration (HT) for figure-eight loops; and flood variability (FV) for complex loops. By combining MEMD-based scale decomposition with objective K-means event classification and HI quantification, we provide a transferable, scale-aware framework for diagnosing sediment transport regimes in semiarid basins. (c) 2025 International Research and Training Center on Erosion and Sedimentation, China Institute of Water Resources and Hydropower Research and Tsinghua University. Publishing services by Elsevier B.V. on behalf of KeAi Communications Co. Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).