In optical remote sensing, relationships used to link radiometric data acquired by satellites and biophysical quantities of vegetation are generally established through semi-empirical relationships or inversion of radiative transfer models. The inversion method of DART model is based on the use of pre-calculated tables: simulations involve a wide set of input parameters. An interpolation procedure, coupled with an analytical model, can recreate a simulation for any values of the input parameters. Inversion is then achieved by minimizing the cost function representing the error between the measured reflectance (satellite images) and reflectance data model. In this paper, we propose an improvement of the inversion method, comparing four parametric models (RPV, MRPV, HAPKE and ESTEVE). To assess their ability to represent the reflectance simulated by DART, we compared the mean square error (RMSE) between the simulated reflectance and those obtained by the different models. The improved MRPV model proved to be more robust to effectively represent reflectance DART. At the end, we applied the inversion to estimate biophysical parameters (leaf area index, crown coverage) of the Fontainebleau forest (France) from a SPOT image. This application allowed us to perform the validation of inversion programs that we have developed and to illustrate the ability to get maps of biophysical parameters that are very useful for modeling the functioning of the forest.