Sub-Canopy Topography Estimation With Multibaseline Pol-InSAR Data: A RELAX-Based Solution

Abstract

New opportunities for the radar remote sensing of forests are arising from L-band and P-band SAR systems given their capability of penetrating deep into volumes. A key objective is estimating the sub-canopy topography, which finds a primary application in the derivation and/or interpretation of vertical radar profiles of the vegetation. Noteworthy, the systematic and reliable vertical forest distribution estimation is one of the essential goals of the German mission proposal TanDEM-L consisting of two cooperating L-band space borne radars flying in close formation. Concerning the ground height estimation problem, a significant processing advance has been achieved in the last decade by coherently combining SAR data acquired in polarization and baseline diversity. For instance, already operating with a single baseline, coherent scattering models can be related to the polarimetric interferometric (Pol-InSAR) complex coherences in order to retrieve the forest height, the extinction coefficient and the sub-canopy topography [1]. In parallel to the single baseline Pol-InSAR, different strategies were investigated. In particular, SAR Tomography (Tomo-SAR) [2-3] and its polarimetric version PolTomo-SAR [4] are experimental techniques which in the last years demonstrated their potential in the 3-D analysis of volumetric scenarios. Tomo-SAR is a multibaseline (MB) extension of conventional cross-track SAR interferometry, employing many passes over the same area. Differently from conventional interferometry, which can only furnish a measure of the terrain topography, (Pol)Tomo-SAR can resolve multiple scatterers at different heights in each given range-azimuth cell, and outputs a continuous profile along the height dimension. Up to now, the performance of Tomo-SAR techniques (parametric and not) for ground topography estimation has been quantitatively assessed mostly with P-band data, proving the possibility to reach a precision in the order of magnitude of 1 m (see e.g. [3, 5]). This work proposes the use and investigates the performance of an iterative ground topography estimation technique employing Pol-InSAR MB data, with particular reference to L-band acquisitions. More in detail, the proposed algorithm aims at separating the ground component (reasonably modeled as a point-like scatterer) from the dominant spatial spectral component of the canopy by means of a superresolution RELAX-based iteration [6-7]. For each iteration, an objective function is optimized with respect to one single height, thus the global algorithm implementation results simple and fast. It is worth remarking that in the signal processing literature it has been demonstrated that while RELAX is asymptotically statistically efficient when dealing with point-like scatterers, it is also robust to a model mismatch [6], which in the specific case happens with the canopy height-extended scatterer. Moreover, in contrast with MB covariance matching techniques like the ones in [5], the ground topography is estimated without making any particular hypothesis on the canopy coherence model, thus without the need of estimating its parameters. In other words, ground and canopy estimations are decoupled. The accuracy in ground topography retrieval is evaluated quantitatively with real data experiments by processing a DLR E-SAR L-band dataset over the Traunstein temperate forest. The dataset consists of 5 fully polarimetric SAR images acquired in June 2008 with a time span of 1 hour, hence with limited temporal decorrelation effects. The maximum baseline measures 20 m. A digital terrain model acquired through LiDAR measurements is also available, and it is used as a benchmark in the performance analysis. A comparison will be carried out of the accuracy obtained by combining single and full polarization MB data. Different baseline distributions, obtained by thinning the original one, will also be considered. References [1] S. R. Cloude, K. Papathanassiou, “Polarimetric SAR Interferometry,” IEEE Trans. on Geoscience and Remote Sensing, vol. 36, 1998. [2] A. Reigber, A.Moreira, “First Demonstration of Airborne SAR Tomography Using Multibaseline L-Band Data,” IEEE Trans. on Geoscience and Remote Sensing, vol. 38, 2000. [3] F. Lombardini, M. Pardini, “Experiments of Tomography-Based SAR Techniques with P-Band Polarimetric Data”, Proc. of 2009 ESA PolInSAR Workshop. [4] S. Sauer, F. Kugler, et al., “Polarimetric Decompositions Applied to 3D SAR Images of Forested Terrains,” Proc. of EUSAR 2010. [5] S. Tebaldini, “Single and Multipolarimetric SAR Tomography of Forested Areas: A Parametric Approach,” IEEE Trans. on Geoscience and Remote Sensing, vol. 48, 2010. [6] M. Pardini, “Advances and Experiments of Tomographic SAR Imaging for the Analysis of Complex Scenarios”. PhD thesis, University of Pisa, 2010. [7] J. Li, P. Stoica, “Efficient Mixed-Spectrum Estimation With Application to Target Feature Extraction,” Proc. of 1996 ASILOMAR Conference.