It is well-known that a typical Reynolds-averaged Navier-Stokes (RANS) model is limited in predicting separated flows. To correct a conventional RANS model, an adaptive correction method is developed in this study. The current adaptive method relies on a RANS solution to identify separated flow regimes. The model correction is adopted from the field inversion process which optimizes a given RANS model for a better prediction. The Spalart-Allmaras (SA) model is chosen in this study for the adaptive correction (i.e., adaptive-field inversion). The adaptive correction method is tested for flows around an axisymmetric base body where massive flow separation occurs at the blunt base. The nominal field inversion (i.e. full-field inversion) takes the whole computational domain for optimizing a RANS model. The full-field inversion often modifies well-attached flow in order to satisfy a given optimization goal. In order to prohibit the model correction in well-attached flows, a zonal-field inversion approach, which specifies separated flow regimes manually, is developed as an intermediate step. The zonal approach successfully eliminates the unwanted modification of the SA model in the upstream boundary layer. The correction is applied to the production term of the SA equation. It is straightforward to divide separated flow fields from the attached flow fields here because of the blunt base in the current geometry. An adaptive method is developed because the zonal approach is limited for a complex geometry where the flow separation point is hardly known before the simulation. The current adaptive function fa is developed from the delayed function fd of the delayed-detached-eddy simulation method. The delayed function fd was designed to identify regimes where the large-eddy simulation mode is active. In the current study, RANS simulation is conducted in the whole computational domain. As such, the function fd needs to be modified to identify separated flow regimes. The detailed formulation of the adaptive function fa will be discussed in the presentation. The current adaptive-field inversion maintains the original SA model in the upstream well-attached flow and, at the same time, corrects the SA model in the separated flow next to the base. The pressure distribution on the base is well-matched to the experimental measurement. Detailed flow conditions and field inversion approaches (both zonal and adaptive methods) will be discussed in the presentation.