In a novel application of the temporal numerical simulation, an investigation ofturbulence modeling techniques is carried for the turbulent wall jet over aconvex surface (Coanda wall jet.) The simultaneous presence of multipleinstability mechanisms and the interaction with the turbulence dynamics at thesolid boundary produces a unique combination of different large turbulentcoherent structures that constitutes both a consistent challenge for numericalsimulations and an ideal test bed for turbulence models. The Temporal Direct Numerical Simulation (TDNS) of the Coanda wall jetrestricts the focus from the global turbulent Coanda wall jet to a smaller, localportion of the flow and offers a wide array of advantages to the present work. Inparticular, the size of the computational domain can be arbitrarily chosen inboth the spanwise and the streamwise directions. This allows to either suppressor enhance individual physical mechanisms and, consequently, to selectivelyreproduce different large coherent structures within the local flow. In the firstpart, temporal numerical simulations are employed to reproduce four differentflow scenarios of the local Coanda wall jet with a level of numerical resolutionthat, because of the reduced size of the computational domain, cannot be matchedby standard DNS of the entire physical flow (spatial DNS, or SDNS.)The TDNS of these four flow scenarios are then used in the second part for ana--posteriori analysis of different turbulence models in order to addresscommon shortcomings shown by Hybrid Turbulence Models (HTM). For each flowscenario, the turbulent flow field is deliberately decomposed in resolved andunresolved flows by the application of different filters in space correspondingto different grid resolution. The behavior of turbulence models can be reproducedfrom the resolved flow and compared to theturbulent stress tensor directlycalculated from the unresolved part of the flow field. Starting from the RANSlimit, turbulence models with different levels of complexity are studied. Successively, the performance of these models is analyzed at intermediatenumerical resolutions between RANS, LES, and DNS. Finally, an improvedformulation of the Flow Simulation Methodology (FSM) is proposed.