In this paper, we present a local Immersed Boundary Method (IBM) to deal with geometrically complex details on unstructured curvilinear meshes for RANS simulations of aerodynamic configurations. This IBM approach consists in forcing the solution at some discretization points close to the obstacle [3]. In the vicinity of the geometrical detail, the mesh may not be well-resolved to capture properly the boundary layer. For instance, the curvilinear mesh in the vicinity of an ice horn on a smooth profile may be too coarse either to capture the geometry or the flow physics locally. For this reason, we propose to adapt the original mesh to the geometrical details and to represent the boundary layer accurately. As this hierarchical mesh adaptation is isotropic, it would be too expensive to adapt the mesh such that the height of the first cell close to the geometrical detail corresponds to y+ ≃ 1 for high Reynolds number flows. Thus, a wall function is applied at IBM forced points to model locally the boundary layer. According to previous studies [2], the mesh resolution targeted by the adaptation in the vicinity of the immersed obstacle corresponds to a y+ ≃ 100 to 300. This approach has been applied to an iced wing profile, where the original mesh is a wallresolved hexahedral mesh around the smooth profile and the geometry of the ice horn is provided. IBM preprocessing is achieved with Cassiop´ee[1] modules and the RANS simulations are performed with CODA flow solver (CFD ONERA-DLR-AIRBUS compressible Navier-Stokes solver)[4]. The ice shape is taken into account as an immersed obstacle. The pressure coefficient distribution obtained by the IBM approach matches with a reference solution on a body-fitted mesh. The IBM approach predicts the position of the peaks of the skin friction distribution but underestimates their intensity. In the final paper, additional validations on 2D configurations and first results of a 3D configuration will be presented.