Numerical models of soil freezing and thawing are being increasingly used in, e.g., agriculture, forestry, ecology and civil engineering. This study was conducted to 1) elucidate the sensitivity in simulation output to the variability of model parameters for the hydrodynamic model Hydrus-1D and 2) investigate how two operational considerations in the model setup, groundwater level and subgrade material (soil texture), affect indicators of road accessibility in northern Sweden. The analysis was carried out by applying the generalized likelihood uncertainty estimation (GLUE) procedure when simulating laboratory measurements of freezing cylinders and by a more conventional sensitivity analysis, varying one parameter at a time, using road surface temperatures measured during nearly 1 year as upper boundary condition. For the simulation of the laboratory experiment, it was found that, although the thermal conductivity scaling factor, λ f, and the convective heat transfer coefficient, h c, most strongly affected the output, no parameter was redundant for the given problem. The frost depth was most sensitive to changes in λ f and h c, while the water content in the unfrozen zone was most sensitive to changes in the hydraulic conductivity impedence parameter Ω. For the 1-year road simulation, the frost depth was larger for sand than for the loam and silt subgrades; the thawing period was shortest for sand and longest for the silt subgrade; and the silt subgrade allowed for the largest frost-induced upward water flow. Thus, among the subgrades studied, roads built on silt show the potential of being most frost-susceptible as a consequence of having the largest elevated water content in combination with the longest time of thawing. The study performed indicates that the model can provide information of interest from an operational perspective, allowing for local predictions important in the road construction and maintenance process.