<p>In coastal ocean modeling, the use of terrain following σ-coordinates is immensely popular but it comes at a cost in the form of baroclinic pressure gradient error (PGE) near steep slopes and shelf breaks, resulting in spurious diapycnal mixing and a loss of stratification. The PGE is problematic in fjords and deep estuaries and is commonly addressed through bathymetric smoothing. This compromise preserves stratification but renders a loss in fidelity of water depths, and in turn a loss in realism, including both qualitative representation of physical processes and quantitative accuracy of predicted currents and temperature. Here we assess the feasibility of eliminating the need for smoothing, using the Localized Sigma Coordinates with Shaved Cell (LSC²) formulation and the SCHISM model over a fjord-like deep estuary, the Salish Sea in the Pacific Northwest, including U.S. and Canadian waters. We upgraded the existing FVCOM based Salish Sea model from 10 σ-layers to a maximum of 25-LSC² layers and expanded the domain 134&#xa0;km further offshore from the shelf break. The model was re-calibrated, and the skill assessed to confirm no drop-off in performance in reproducing stratification and expected improvement in predictions of currents and temperature from the upgrade to original unsmoothed bathymetry. With the same lateral resolution and original bathymetry, stable stratified conditions in the Salish Sea were reproduced at a level of skill, comparable to prior smoothed bathymetry model version. Major known circulation features and distinguishing characteristics separating sub-basins into fjords, partially mixed, and costal-plain behaviors were reproduced. Most importantly, a significantly improved performance in predicting currents was observed in regions with steep slopes and narrow channels. In the shallow intertidal regions, improved performance was noted for currents as well as temperature confirming the ability to conduct cross-scale simulation in a fjord environment.</p>