Modeling salinization of wells in semi-confined aquifers with a fresh-salt interface presents many challenges. Accurately modeling of upconing and the salinity of extracted groundwater under variable-density conditions requires a fine spatial and temporal discretization, which is computationally prohibitive at a regional scale. A common practical approach to address this challenge is to use larger-scale models to derive boundary conditions for locally refined models. In this way the effects of regional topography or seasonality can be accounted for in the local model. Transfering the heads and fluxes calculated with the regional model to the smaller submodel is not a trivial task, especially when the regional groundwater models do not include salt transport and variable density flow. This research aims to answer the following questions. How do you determine the appropriate dimensions and resolution of the locally refined model? What errors are introduced when boundary conditions derived from a constant-density regional groundwater model are applied to a local variable-density model? How to account for the density effects that occur in the local model but not in the regional model? To investigate these questions a synthetic regional Modflow 6 model is used. The regional model includes hydrological processes such as groundwater recharge, ground- and surface water interaction, wells and variable density flow. Extraction wells are simulated at a number of key locations in the model that differ in, for example, the depth to the interface and the geohydrological layer structure. The effect of neglecting variable-density flow is investigated using this model. A constant-density version of this model is also created. Locally refined models are constructed at the well locations. The effects of model size, resolution and boundary conditions on the simulated extraction water salinity and strategies for translating the heads and fluxes from the density-invariant version of the regional model to boundary conditions for the local models are investigated. The lessons learned will be presented as practical guidelines for building submodels for well salinization based on models of regional flow systems.