Soil erosion produces enormous amounts of sediment, carbon, and nutrient fluxes from land to rivers, thus playing crucial roles in global biogeochemical cycles, water quality and food security. To predict soil erosion in the context of climate and land use change, we explicitly represent the effects of cropland management actions (i.e., irrigation, conserved agriculture, and crop residue management) and geological factors (i.e., lithology and glacier) in the soil erosion model of Energy Exascale Earth System Model (E3SM). The model is calibrated consistently using a global-scale regionalized parameter calibration method. The spatial variability of the modeled soil erosion across vegetation, climate and soil types is consistent with the benchmark data produced by Revised Universal Soil Loss Equation (RUSLE). Comparison of our model and RUSLE with independent soil erosion data shows that our model reduces model biases in 59% of the observations, with 53% of the bias reduction exceeding 50%. This improvement mainly results from a better representation of the topographic influence on soil erosion. Our simulations indicate that soil erosion in cropland can be reduced by over 25% with wide adoption of conserved agriculture but irrigation may significantly increase soil erosion in cropland over India and China. For upland sediment flux, our model is consistent with the WBMsed benchmark data in inter-basin variability but it may be more skillful in simulating intra-basin variability because soil erosion and sediment flux are more closely coupled in our model. The developed model may enable more realistic predictions of soil erosion and river sediment dynamics in response to environmental changes.