Vegetation structure estimation from SRTM coherence data: correction of systematic artifacts

The Shuttle Radar Topography Mission (SRTM), sponsored by NASA and the National Geospatial-intelligence Agency (NGA), flew in 2000 on the Space Shuttle Endeavour and acquired the only global-scale space borne single-pass Interferometric SAR (InSAR) data set in existence. One C-band Synthetic Aperture Radar (SAR) was in the cargo bay of the Shuttle, while 60 meters away, attached to a retractable mast, was another C-band antenna that could receive the reflected signal as well. In order to increase the swath width, SRTM operated in a ScanSAR mode, in which bursts of pulses were transmitted after pointing the antenna beam to one of 4 possible subswaths. For two subswaths, the transmitted pulses were horizontally polarized, while for the other two subswaths the pulses were vertically polarized. By the end of the mission, for every location on Earth between ± 60 degrees, SRTM acquired crossing paths of InSAR horizontally and vertically polarized data, with incidence angles between 20 and 60 degrees. The complex imagery from each antenna was used to form interferograms with a 60 m baseline. From these interferograms, the SRTM Digital Elevation Model (DEM) was created, and is currently in wide use for many applications. Several previous studies have shown that the magnitude of the interferometric coherence has a well-defined relationship to the structure of vegetation that may lie within the image swath. Unfortunately, for repeat-track InSAR systems, decorrelation associated with weather events and transitory changes in vegetated areas can overwhelm this signature, and prevents this technique from being successfully applied routinely to widely available repeat-pass InSAR data. However, since SRTM was a single-pass InSAR mission with a desirable separation between antennas, some characteristics of vegetation structure should be observable within the SRTM correlation data, if there is vegetation present in the image swath. Unfortunately, the SRTM coherence data must first be corrected. Due to burst-to-burst variations in the Signal to Noise Ratio (SNR) component of the coherence, the coherence data have burst-to-burst banding visible throughout much of the imagery. During mission processing, the spatially varying nature of the SNR made it difficult to find a universal correction procedure. While this banding did not prevent attaining the desired accuracy for the DEM, it proved problematic to produce 30-100 m resolution measurements of the volumetric correlation, which is the component of the observed coherence that is related to vegetation structure. Image-processing techniques combined with appropriate interpolation between minimally affected pixels could be used to filter out some of the variations of the banding in the imagery (at the expense of resolution). With a less demanding resolution requirement, NASA has funded a reexamination of the prospects of generating a volumetric correlation product that may be linked to vegetation structure in a robust and unified manner. This paper was written at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration.