Sediment transport is inherently multi-scale and nonlinear, driven by interactions among meteorological, hydrological and ecological processes. However, the scale-dependent relationships between sediment yield and its potential drivers remain poorly understood since multiple underlying mechanisms overlap at the observational scale. To address this gap, this study applies multivariate empirical mode decomposition (MEMD) to identify the scale-specific controls on sediment production. Annual sediment yield and five influencing factors-runoff, temperature, precipitation, potential evapotranspiration, and the normalized difference vegetation index-were compiled for six sub-basins of the Yellow River Basin during 1985-2017. Each variable was decomposed into intrinsic mode functions (IMFs) and a residual component, allowing scale-specific sediment-environment linkages to be identified. The results showed that sediment yield exhibited significant associations at specific scales, including cases where no significant relationship was observed at the observational scale. These controlling relationships were strongly scale-dependent and spatially heterogeneous, with several predictors exerting greater influence at longer time scales. Scale-specific drivers reproduced sediment yield effectively for individual IMFs and residuals, while sediment yield at the observational scale was accurately reconstructed by summing all scale-based predictions. Runoff and precipitation emerged as the dominant predictors in most sub-basins, whereas basin-specific differences were observed for temperature and potential evapotranspiration. MEMD-based modeling substantially improved sediment prediction accuracy (R2 = 0.71-0.93) compared with models using undecomposed data (R2 = 0.37-0.86). These results demonstrate that sediment-environment interactions cannot be fully captured at a single time scale and highlight the potential of MEMD for disentangling multiscale sediment dynamics and improving sediment-yield predictions in large river basins.