In this work, a volume-filtered Eulerian-Lagrangian (VFEL) approach is employed to evaluate the model fidelity required to accurately predict the dynamics of sub-aqueous sedimentary flows. The VFEL approach allows for interphase exchange terms (e.g., volume fraction and drag) to be computed on a length scale independent of the grid resolution, enabling a comprehensive analysis of the errors associated with numerical discretization and physical models on bed height, particle flux, and bed wavelength. Simulations of initially flat granular beds under both laminar and turbulent shear flow are carried out in featureless and pattern forming regimes. Resulting errors are evaluated by comparing VFEL to the direct numerical simulation (DNS) data of [1,2]. Grid refinement demonstrates convergence of the bed height and particle flux in all of the cases considered. However, traditional drag laws based on local volume fraction and particle Reynolds number lead to systematic errors in the featureless laminar regime. A stochastic drag law that accounts for the variance in local particle configuration provides improvement of the steady state statistics. The use of a stochastic drag law was found to be less important when the rise in bed height is minimal. Finally, the VFEL approach with stochastic drag is used in a large-eddy simulation framework to simulate pattern formation in turbulent shear flow at relatively high Galileo numbers. (C) 2019 Elsevier Ltd. All rights reserved.