We propose in this paper an hybrid method combining a finite-volume method and a semi-lagrangian particle method to study the advection-diffusion of scalars in turbulent flows and at high Schmidt numbers. A particular focus is made on the coupling of the two methods in the context of high performance computing. Results are given concerning the accuracy and the scalability of the method. This method is then used to study differential diffusion on two canonical cases: Homogeneous Isotropic Turbulence and a jet flow. We first characterize differential diffusion in terms of spectral distribution. We then use the Reynolds decomposition to bring out the different mechanisms involved in the energy budget of the scalar and we analyze their spatial distribution.