At different time scales, the potential factors influencing changes in sediment transport vary. Because the processes of each impact factor overlaps on different time scales, it is challenging to evaluate the complex multi-scale relationships between monthly sediment load and its potential impact factors. The objective of this study was to investigate the scale-specific main factors influencing monthly sediment loads using the multivariate empirical mode decomposition (MEMD) method. Monthly sediment loads and five potential influencing factors (runoff, precipitation, air temperature, potential evapotranspiration, and enhanced vegetation index) during 2003–2017 were collected in the Wujiang karst watershed of southwest China. The MEMD method was used to decompose the temporal series of monthly sediment load into seven intrinsic mode functions (IMF) and residuals using a Hilbert transform. The sum of variance contribution rates of IMF1 (3.2 months), IMF2 (5.5 months) and IMF3 (12.1 months) was more than 90%. Although temperature and potential evapotranspiration significantly affect monthly sediment loads at the observation scale, no significant relationships were observed between them at some specific scales after MEMD. The accuracy of the prediction model after MEMD was better than that of the prediction model using original data. Runoff and precipitation were important predictors in the prediction model. This study shows the advantages of the MEMD method for analyzing non-stationary and nonlinear hydrological processes, and this useful tool is recommended for application in other karst watersheds.