Soil water dynamics are central in linking and regulating natural cycles in ecohydrology, however, mathematical representation of soil water processes in models is challenging given the complexity of these interactions. To assess the impacts of soil water simulation approaches on various model outputs, the Soil and Water Assessment Tool was modified to accommodate an alternative soil water percolation method and tested at two geographically and climatically distinct, instrumented watersheds in the United States. Soil water was evaluated at the site scale via measured observations, and hydrologic and biophysical outputs were analysed at the watershed scale. Results demonstrated an improved Kling–Gupta Efficiency of up to 0.3 and a reduction in percent bias from 5 to 25% at the site scale, when soil water percolation was changed from a threshold, bucket-based approach to an alternative approach based on variable hydraulic conductivity. The primary difference between the approaches was attributed to the ability to simulate soil water content above field capacity for successive days; however, regardless of the approach, a lack of site-specific characterization of soil properties by the soils database at the site scale was found to severely limit the analysis. Differences in approach led to a regime shift in percolation from a few, high magnitude events to frequent, low magnitude events. At the watershed scale, the variable hydraulic conductivity-based approach reduced average annual percolation by 20–50 mm, directly impacting the water balance and subsequently biophysical predictions. For instance, annual denitrification increased by 14–24 kg/ha for the new approach. Overall, the study demonstrates the need for continued efforts to enhance soil water model representation for improving biophysical process simulations.