On the use of an "optimal database" for radar forest observation

Abstract:
In this paper, we apply a fast method using an optimal database to estimate the impact of some radar sensor characteristics on three quantities (called "outputs"): the polarimetric scattering coefficient, the interferometric height and the total attenuation of the forested area under study. We consider two characteristics: the radar frequency and the incidence angle - called the "inputs" of the model. A variation domain for both of them is defined, leading to the so-called input space. Then, simulations are performed sequentially, controlled by a sophisticated strategy in order to build an "optimal database". The latter is a set of corresponding input-output data pairs (samples), so that the output data are spread out as well as possible. In other words, the "output space" is filled by the samples of the optimal database completely and uniformly. The benefit of such an optimal database is multiple. First, the time-consuming numerical simulations can be replaced by a fast interpolation based on the samples stored in the database. Second, the structure of the optimal database reflects meta-information about the studied problem, helping us to see and understand better the behavior of the physics behind our model. This work might be a possible first step to reveal relationships between the observable radar characteristics and e.g., the true mean height of the forest.