There is a recent strong activity in the study of mixing, dispersion and transport of oceanic properties from the Lagrangian perspective. An important Lagrangian tool which is becoming widely used in oceanography is that of the Finite-Size Lyapunov Exponents (FSLEs). They are a local measure of particles dispersion which, most importantly, serves to characterize coherent structures in the turbulent oceanic flow. In this work we show the FLSE capacity for the identification of Lagrangian structures and their mixing activity at different spatial scales, from coastal to global scales. At the submesoscale, we analyze Palma Bay using data from a ROMS model at 300m of spatial resolution. At the mesoscale we use velocity data from the DieCast model in the Balearic Sea, and for the global distribution we use the world ocean data from JAMSTECH numerical model (Earth Simulator). A seasonal analysis and a comparative study between different regions are performed. Although mathematically appealing, it is rather unclear how robust are FSLE analyses when confronted to real data, that is, data affected by noise and with limited scale sampling. In this study, we analyze the effect of finite scale samplings and of diverse types of noise on FSLE diagnostics. Both effects can be accounted to determine which part of the diagnostics is reliable. Most importantly, scale dependence of FSLE reveals the emergence of a cascade-like structure in oceanic flows, which can be used to improve diagnostics and to better understand ocean dynamics.