To make full use of the increased possibilities of imaging spectroscopy (compared with the traditional multispectral instruments) for remote sensing of vegetation canopies, physically-based models should be used. The problem of retrieving the large number of model parameters from remotely sensed reflectance data is an ill-posed and underdetermined one. The physically-based spectral invariants approach may, in some cases, seem a lucrative alternative. However, the various formulations presented in literature are sometimes difficult to compare qualitatively or quantitatively. To develop a robust spectral-invariant based algorithm for vegetation remote sensing, empirical, mathematical and physical understanding of the problem has to be reached. We present connections between the photon recollision probability and the largest eigenvalue of the radiative transfer equation. Based on simple mathematical principles, the basic requirements set by the remote sensing process to a successful spectral invariant theory are presented.