FIRST EXPERIMENTS OF SECTOR INTERPOLATED SAR TOMOGRAPHY Fabrizio Lombardini, Matteo Pardini Dept. of Information Engineering – University of Pisa Via G. Caruso, 16 – 56122 – Pisa – Italy {f.lombardini, matteo.pardini}@iet.unipi.it 1. INTRODUCTION Synthetic aperture radar 3-D Tomography (Tomo-SAR) is an experimental multibaseline (MB) extension of conventional cross-track SAR interferometry employing many (on the order of ten) passes over the same area. Through the coherent combination of MB data at the complex (i.e. amplitude and phase) data level, Tomo-SAR can separate multiple scatterers at different heights in each given range-azimuth cell, and produces a continuous reflectivity profiling along the height direction ( h direction). Generally speaking, Tomo-SAR exploits an aperture in the height-range plane constituted by the MB geometry to get full 3-D imaging through elevation beam forming (spatial spectral estimation). Applications of Tomo-SAR have been found for biomass estimation, forest classification, tree and building height estimation, and other geophysical parameter extraction problems [1]. However, in practice possible temporal decorrelation of the scattering and problems of cost prevent from using a large number of passes, while navigation/orbital considerations do not allow obtaining ideal planned uniformly spaced parallel flight tracks. As a consequence, the point spread function (PSF) along the h direction is distorted. In particular, the non-uniform baseline (spatial) sampling causes bad h-imaging quality results with the classical Fourier-based focusing in terms of contrast and ambiguities, as anomalous side and quasi-grating lobes affect the PSF. Alternatively to the Fourier-based processing, a regularized inversion approach to Tomo- SAR was first introduced in [2], whereas in [3] an adaptive beamforming technique was proposed and tested, providing strong height super-resolution and sidelobe suppression capabilities. As a drawback, the adaptive processing is sensitive to residual data miscalibration, and it exhibits radiometric non-linearities. Also, to operate it needs a spatial coherent multilooking, i.e. it does not operate at full horizontal resolution. More recently, other solutions have been also proposed and tested for the 3-D focusing and parameter extraction [4]-[8]. In a different class of tomographic processors, the MB data are pre-processed by filling the gaps in the available baseline distribution. Under the simplifying assumption that a single height backscattering contribution is dominant in the range-azimuth cell, a simple gap filling algorithm has been proposed in [1]. However, unsatisfactory results are obtained in scenarios comprising more than one point-like spaced or extended scatterer [9], because of the assumption mismatch. An extension of the interpolator in [1] to more general scenarios has