Estimation of Ground Topography in Fortested Terrain by Means of Polarimetric SAR Interferometry

Kurzfassung

Forest covers approximately 30% of the Earth's solid surface area, with a mean tree height of about 20m. Any attempt to provide global surface mapping based on SAR Interferometry (InSAR) is affected by a bias error due to vegetation. The magnitude of this error depends on the used frequency and on the forest extinction; and may range up to the mean tree height. The evaluation of volume decorrelation effects in multi-baseline InSAR data has demonstrated that there is no conventional frequency (from X- up to P-band) and/or polarisation able to “see” only the ground under a vegetation layer without being affected by any volume (vegetation) scattering contribution. In consequence all Digital Elevation Models (DEM’s) generated by means of InSAR are affected by a more or less significant vegetation bias. The correction of this inherent vegetation bias - always present in conventional interferometric data - and the estimation of the underlying ground topography is an essential improvement of the topographic information provided by SAR interferometry, with great ecological as well as commercial impact. The estimation of the underlying ground topography in forested terrain is inherently related to the problem of separating volume from surface like scatterering contributions in the case of a mixed (surface and volume) scattering scenario. InSAR measurements provide a key element towards the solution of this problem as volumes (where the scattering phase centers of the individual volume scattering elements are distributed over the whole vertical volume extend) and surfaces (with an vertically localised scattering center) have a fundamentally different interferometric behavior. However, the separation of volume from surface scattering centers attempted in different ways by means of multi-baseline InSAR measurements lead – for most conventional implementations - to an ill-posed inversion problem. The coherent combination of (quad- or dual-) polarimetric and interferometric measurements by means of Pol-InSAR leads to a better conditioned inversion problem. Indeed model based separation, associated with the inversion of the Random Volume over Ground (RVoG) model [1], has been successfully demonstrated for a variety of forest and terrain conditions, primarily focused on the evaluation of the volume component (forest height and structure). The proposed methodology for the estimation of the ground topography is based on the geometrical interpretation of the RVoG and do not require the full model inversion. A two-fold ambiguity must be resolved through separate constraints [1-3]. This can be done based on physical considerations but becomes more and more difficult with decreasing InSAR coherence level. However, in contrast to the volume component parameters, the estimation perfromance of the ground topography has not been validated in detail up to now, primarily due to the lack of accurate reference data. In this paper we propose an alternative implementation of the RVoG inversion for estimating underlying ground topography from Pol-InSAR data. The main advantages of the proposed approach are a more robust (by means of parameter estimation) implementation and an unambiguous estimation. The estimation performance the proposed approach is evaluated. For this we estimate ground topography over a variety of forest conditions using repeat-pass Pol-InSAR data acquired by DLR’s airborne E-SAR system at L-band in a repeat-pass InSAR mode. The obtained results are validated against available LIDAR ground DEMs used as reference. Critical effects are discussed: The effect of model validity on the estimation performance is evaluated and simulated; the impact of non-volumetric decorrelation contributuions (as temporal decorrelation) is discussed and optimised system and acquisition scenarios are proposed.