Mixing parameters can be inaccurate in ocean data assimilation systems, even if there is close agreement between observations and mixing parameters in the same modeling system when data are not assimilated. To address this, we investigate whether there are additional observations that can be assimilated by ocean modeling systems to improve their representation of mixing parameters and thereby gain knowledge of the global ocean’s mixing parameters. Observationally-derived diapycnal diffusivities–using a strain-based parameterization of finescale hydrographic structure–are included in the Estimating the Circulation & Climate of the Ocean (ECCO) framework and the GEOS-5 coupled Earth system model to test if adding observational diffusivities can reduce model biases. We find that adjusting ECCO-estimated and GEOS-5-calculated diapycnal diffusivity profiles toward profiles derived from Argo floats using the finescale parameterization improves agreement with independent diapycnal diffusivity profiles inferred from microstructure data. Additionally, for the GEOS-5 hindcast, agreement with observed mixed layer depths and temperature/salinity/stratification (i.e., hydrographic) fields improves. Dynamic adjustments arise when we make this substitution in GEOS-5, causing the model’s hydrographic changes. Adjoint model-based sensitivity analyses suggest that the assimilation of dissolved oxygen concentrations in future ECCO assimilation efforts would improve estimates of the diapycnal diffusivity field. Observationally-derived products for horizontal mixing need to be validated before conclusions can be drawn about them through similar analyses.