ESA > Join & Share > Technology Projects > COISP Project

COISP Project

Project Title   Preparation of Interferometric SAR Processor
Project Acronym   COISP
Contractor(s)   Technical University of Denmark (Denmark)


Project Context
          How it works



In a previous project, an interferometric post-processing (IPP) software package was developed at the Technical University of Denmark (DTU). The IPP now takes ERS and ENVISAT raw data or SLCs as input, generates digital elevation models (DEMs) and displacement maps in geographical coordinates, and attaches maps of predicted error standard deviations to the data. The system also allows ingestion of interferograms, and can work as a stand-alone error predictor for other systems. The modules are fully chainable and work with a minimum of user intervention.

The tiled output format in geographical coordinates was chosen to support large area processing, using a database approach. Other key characteristics of the IPP are its error prediction and the automatic geo-coding based on precision orbit data. The system is used at DTU, at ESRIN, and preparation for inclusion in the KEO framework is on-going.



The objectives of COISP were to broaden the application field of the IPP, improve robustness, and improve the error prediction capabilities. The following specific tasks were identified:

  Image Provision of an ERS SLCI interface
  Image Implementation of a strip facility
  Image Development of an atmosphere and GCP error prediction function
  Image Robustness improvement
  Image Performance optimization

The objective of the first task was to establish an ERS.SAR.SLCI interface. This interface was implemented and now allows input of SLCs and thus utilization of the huge ERS archive for small projects where use of 100 km by 100 km SLCs is feasible. As a spin-off, support for other sensors with data provided in the CEOS format was implemented.

The objective of the second task was to develop the functionality for processing long strips of raw data (more than a 100 km standard scene) from ERS. This includes both concatenation, cleaning and focusing of ERS.SAR.RAW data and (re ) formatting of raw down-link data on DLTs followed by a traditional SAR processing and data annotation.

The work of the third task resulted in an error prediction framework for single interferogram DEMs, single interferogram displacement maps, and double difference products, 1. The framework assumes that a tie-point scheme is used for baseline and absolute phase estimation. Its output is the predicted standard deviation of measured height/displacement for each pixel. Also, the framework allows for calculation of relative errors (error correlation between any pair of pixels). However, efficient presentation of such information was not developed. The framework uses models for: ground control point elevation and displacement uncertainty, phase noise (described by the interferometric coherence), atmospheric delays, and phase unwrapping errors. A tuneable closed form model for the atmosphere was developed and described in 2. A first version model for phase-unwrapping uncertainties was described in 1, and this was further developed as described in the project documentation and in the final presentation (see link below).

The fourth task was split into different elements. The first concerned weight assignment for the minimum cost flow phase-unwrapping algorithm. It was found that a small weight (about 1/1000 of the maximum weight) provided a good balance between reasonable unwrapping time (higher weights reduce unwrapping time) and more reliable phase unwrapping (lower weights increase reliability). The second concerned methods for removing possibly unreliable unwrapped loosely connected regions and bridges. Finally, the phase unwrapping error model was improved and tuned.

The fifth task on performance optimization was split into two. The first was to port the software to Linux in to get easy access to the continuously increasing cheap computing power from PCs. The second was to reduce processing time. T he work included: a) serial optimization, b) selection of optimal compiler, libraries and compiler flags, and c) parallelization. For the selected test scenario a factor 10 improvement was achieved on a pc with 4 processors.

1 Mohr and Merryman Boncori, "An Error Prediction Framework for Interferometric SAR Data", to appear in IEEE Trans. Geosci. Rems. Sens., vol. 46, no. 6, June 2008.  
2 Merryman Boncori and Mohr, "A Tunable Closed Form Model for the Structure Function of Tropospheric Delay", to appear in IEEE Geosci. Rems. Sens. Letters, vol. 5, no. 2, 2008.  



For large area mapping, interferometric measurements from many strips have to be combined into a single output product, 3. In some areas it might also be advantageous to use information from other data sources, either due to a superior data quality or in order to fill holes. Keywords for this task could be Kriging and draping.

Also, the many strip intersection areas can potentially be used for calibration, possibly in conjunction with information from other data sources. Here the coupling in InSAR products between elevation and across track position has to be taken into account, i.e. if the elevation of an area is adjusted, the image also has to be shifted horizontally. This is somewhat incompatible with the concept of combining InSAR measurements from different strips. For example, in order to compare the elevations derived from an ascending and a descending orbit coverage of an area, the elevations have to be georeferenced. Another issue, which is important for non-stationary surfaces such as glaciers, is that the displacement measured by an interferometric SAR system is dependent on the line-of-sight direction. This implies that unless the InSAR images are acquired from the same track, the motion components in the overlap areas do not have to be equal as is the case for the elevation data.

The core of the strip calibration task thus includes setup and solution of calibration equations, adjustment of the elevations and displacements, and a horizontal resampling corresponding to the elevation adjustments. The combination of different strips requires an initial geocoding of reasonable accuracy. This can be accomplished with a conventional inversion of interferogram phase to elevation and/or displacement. If the inversion does not include estimation of elevation, an external DEM is required.

The above considerations have resulted in a design concept with three processing elements:

  Image Geophysical Inversion Mode (GIM).
  Image Data Adjustment Mode (DAM).
  Image Fusion Mode (FUM).

The GIM is a conventional interferometric inversion tool. It uses one or two interferograms (all from the same track) as input, and produces one output product with elevation (either derived interferometrically or from an external DEM) and/or the line-of-sight displacement. The geophysical measurements are calibrated using ground control points (GCPs) and/or an external DEM. The output is resample to a geographical coordinate system in order to allow calibration by comparison with other strips in the (optional) subsequent processing steps. The output product also includes an error model, which e.g. provides a relation between an elevation error and a horizontal shift. The output, which are provided in a tiled format, is denoted a Geophysical Product level 1 (GPP-1). Typical tile sizes are 1º by 1º.

The DAM uses many GPP-1 products as input, which is calibrated, and all output still being GPP-1 products. Additional input are GCPs and an external DEM. This component has not yet been implemented.

The FUM fuses many GPP-1 products and possibly an external DEM into one final output, denoted a geophysical product level 2 (GPP-2). The GPP-2 product is also a tiled product.

Since standard archive data are single look complex images (SLCs), (or raw data), the core system also includes an interferogram formation module (IFF).

In addition, a front-end SAR processor was integrated to a focusing module (FOC) and the previous project included development of a conversion utility module (CUT). The purpose of the CUT is to "cut" out a user-defined window from the GPP-2 (or GPP-1) database of tiled elevations and displacement measurements.

The architecture of the IPP processing depicted below is believed to be quite unique. To our knowledge, the system closest to the IPP is the German GEMOS system, 4, and the processing system for the NASA Shuttle Radar Topographic Mission (SRTM) 5. However, the technical challenges for SRTM are somewhat different, since it is a single pass system.




3 Mohr, J.J. and Madsen, S.N., “Processing Interferometric ERS-1/2 Tandem data Coast to Coast in Greenland,??? ESA ERS-ENVISAT Symposium, Gothenburg, Sweden, 16–20 October, 2000, (8 pages).
4 Eineder, M., Schättler, B., Hubig, M., Knöpfle, W., Adam, N., and Breit, H., “Operational processing large areas of interferometric SAR data???, In Second International Workshop on ERS SAR Interferometry, `FRINGE99', Liège, Belgium, 10-12 Nov 1999, 6 pages . ESA publication SP-478, 1999.
5 Hensley, S., Rosen, P., and Gurrola, E. “The SRTM topographic mapping processor???, In: International Geoscience and Remote Sensing Symposium 2000 (IGARSS’00)???, pp. 1168–1170, 2000.


How it works

The input and output is summarized below:




An example of a processing sequence is given in the final presentation (see link below).



The main project output was an upgraded interferometric processing system and a stand-alone module for error prediction in interferometric DEMs and displacement products calibrated by use of ground control points. The software was installed at ESRIN and presented at a final presentation final presentation on 10th April, 2008.


Contributors to this page: andreadv .

Page last modified on Wednesday 22 of December 2010 15:10:34 CET by andreadv.