Loading...
 
ESA > Join & Share > Technology Projects > PEPSI Project
Print

PEPSI Project

Preparation of an advanced standalone error prediction module for SAR interferometry (PEPSI)





PROJECT AT A GLANCE

In previous projects, an InSAR post-processing (IPP) software was developed at the Technical University of Denmark (DTU) link to COISP project page. The processor included a standalone Error Prediction Module (EPM), applicable in principle to the intermediate output of any classical InSAR processing chain, to provide error estimates (predicted error standard deviation maps) of height and displacement measurements.



OBJECTIVES AND BENEFITS

PEPSI's objectives fall within two broad groups:


InSAR processor upgrade

  • Support TerraSAR-X COSAR (Level 1) format
  • Support COSMO-SkyMed SCS (Level 1) format
  • Support ROI_PAC SLC (Level 1) format
  • Implement DEM assisted co-registration
  • Implement a Stack Coregistration and Interferogram formation module (SCI)


Error Prediction Module enhancement

  • Tropospheric propagation: use mesoscale Numerical Weather Prediction data for error mitigation (stratified delay) and improved error variance estimation (turbulent delay)
  • Phase unwrapping: use available external datasets (external DEM) for error mitigation

APPLICATION AND RESULT EXPECTED

InSAR processor upgrade

The TerraSAR-X SSC product can be directly ingested by the processor's Interferogram Formation (IFF) module, without the need of converting to any intermediate binary format. The software uses the xmlUtil library, based on libxml2. A standalone reader program to dump the contents of the XML annotation file and optionally the image data was also developed.



The COSMO-SkyMed SCS product can be directly ingested by the processor's IFF module, without the need of converting to any intermediate binary format. The software uses the hdf5Util library, based on libhdf5. A standalone reader program to dump the annotations contained in the HDF5 file and optionally the image data was also developed.



The ROI_PAC deskewed SLC format (.slc binary file, .slc.rsc ASCII file and .orrm ASCII file) can be ingested by the processor's IFF module, without the need of converting to any intermediate binary format, after an ASCII parameter (.prm) file in the IFF's Generic SLC format has been created using the provided roipac2iff script.



DEM assisted coregistration was implemented, by allowing the IFF to use a supplied external DEM (ellipsoidal or geoid height) when creating a resampling lookup table. The DEM is used to calculate slant-range shifts, which in turn are used to compute the range spectral filtering centre frequencies, improving the coherence for terrains with a high slope. The computed shifts are also used to flatten the SLC and thus the interferogram phase.



A stack coregistration and interferogram formation module was developed as a first step towards multitemporal InSAR processing techniques. The module coregisters a stack of images acquired from the same ground track, using a redundant network approach, and allows the user to generate single-master or multiple master interferograms, registered to a common geometry.


File not found.


Error Prediction Module enhancement


The processor's Geophysical Inversion Module (GIM) and the standalone Error Prediction Module (EPM) were modified to accept a meteorological product on input, as can be output by a mesoscale NWP model. The product can be of two types: two raster files in a lat/lon grid containing the one-way Zenith Total Delay (ZTD) and the Zenith Wet Delay (ZWD) values respectively; three raster files in a lat/lon grid containing, namely, surface temperature, surface pressure and precipitable water vapour. In both cases an ASCII .grd file (in the processor's GPP 1.0 format) must also be provided to describe the data geometry (e.g. posting, corner coordinates).



The GIM was modified to estimate a linear regression coefficient of ZTD vs. height, for each image acquisition. The latter is then used to compensate the interferometric phase prior to phase unwrapping. The procedure was tested using the output products from the Danish Meteorological Istitute (DMI) operational HIRLAM model (3 x 3 km resolution) and an ASAR dataset of 8 InSAR pairs covering the Geirangerfjord, Norway. The NWP data and the SAR acquisitions were less than 5 mins apart. In most cases the correction based on the HIRLAM model provided a positive improvement of the observed phase residuals.


Image Image



Image



Image



The GIM and EPM were also modified to estimate the parameters of a Stable variogram model from provided or computed ZWD maps for each image acquisition. The variograms are in turn used to predicted the error standard deviation of the measured height and/or displacement, according to the framework of Mohr and Merryman, 2008.



The procedure was tested using the output products from the DMI operational HIRLAM model (3 x 3 km resolution) and an ASAR dataset of 6 InSAR pairs covering the Netherlands, northeastern Germany and Denmark. The NWP data and the SAR acquisitions were less than 5 mins apart. In most cases the estimated experimental variogram parameters were unphysical. In the remaining, the performance of the predictions of the tuned model were not better than those based on the default "off-the-shelf" model, with parameters tuned to average mid latitude conditions.


Image Image
Image Image


Concerning phase unwrapping error mitigation, an approach based on the availability of an external DEM was implemented to improve the reliability of the segmentation mask used to identify consistently unwrapped regions. The approach was tested on an ERS Tandem dataset used to derive a DEM of Disko Bay, in western Greenland.


Image

Contributors to this page: .

Page last modified on Wednesday 09 of January 2013 17:08:45 CET by .