INTRODUCTION

This page describes a set of python scripts that uses ROI_PAC software to create dense sub-pixel offsets of SAR and optical images.

======================

ANNOUNCEMENTS

09/04/2009: "LEA*.001" is now recognized as a valid format for ERS leader files. 09/04/2009: Note that "pixelTrack.py" generates "pixelTrackLog.txt", which records the start step, end step and elapsed time for each run of "pixelTrack.py". This note is included in the pdf. 08/24/2009: "azo.pl" and "azo_merge.pl" have been added to "pixelTrack.tar". This should enable the correct functioning of the "ampcor" step.

======================

BACKGROUND & TECHNICAL DETAILS

ROI_PAC generates offsets using the method of normalized cross correlation (NCC), which computes the offsets in the spatial domain (like the program IMCORR, available here). Another open-source program to generate sub-pixel offsets is called COSI-CORR (available here) and uses both NCC and another algorithm that takes advantage of the Fourier Shift Theorem and computes the offsets in the frequency domain. COSI-CORR requires the use of the GIS package ENVI and the commercial software package IDL, which are usually available for $2000-$3000 to academic users.

The NCC and Fourier methods have different advantages (see, for example the review paper by Brown, 1992). For certain types of data, Sebastien Leprince has found that the Fourier methods produce better results than the NCC methods, although the NCC methods are sometimes more robust for noisy images (e.g., Leprince et al., 2007). We have found that the NCC method can usually match the precision of the Fourier method when the data has been high-pass filtered (e.g., Scambos et al., 1992) before doing the sub-pixel offset tracking, as would be expected (Leprince, personal communication).

The python scripts available for download here are thus complementary to the IMCORR and COSI-CORR packages. COSI-CORR is very user-friendly, and works with a wide range of different types of optical sensors (include aircraft photos). However, COSI-CORR will not currently work with SAR images and requires the purchase of some commercial software, which can be expensive. Our script will work with both SAR data (available at UNAVCO, ASF, or elsewhere) and can go from raw data to a final, georectified set of offsets. While our software will calculate offsets for optical images, our package will not do the required orthorectification, and so this step must be done by different software or by the data provider (for example, product 14 for ASTER). IMCORR will work with both SAR and optical images but requires additional software to create images from raw SAR data, and for orthorectification of optical images.

More information about the theory behind sub-pixel offset tracking is available in these notes from the 2009 UNAVCO short course.

========================

DOWNLOAD & INSTALL

A zipped file containing the python scripts and an instruction manual is available here. The instruction manual includes information on what software packages must be installed in addition to ROI_PAC and MDX.

=========================

TEST DATA

We recommend testing the software with various test datasets to ensure that the programs are correctly working on your computer. Several datasets are available:

* For SAR, we recommend downloading the pair of images from the Hector Mine Earthquake, which were used to in the publications by Simons et al., 2002; and Jonsson et al., 2002. The data are available in the WInSAR archive, track 127, frame 2907, dates 10/20/1999 and 9/15/1999. You can download the DEM you need from the SRTM website -- if you download tile N35W117.hgt.zip and rename it as hector_mine.dem you can use the following hector_mine.dem.rsc file.

* For optical images, the ASTER team has generously made 2 pairs of raw images available for testing at no-cost: a pair of images spanning the 2005 Kashmir earthquake (published in Avouac et al., 2006) and of the Helheim glacier in Greenland (published in Howat et al., 2005). Each ASTER image is available in 2 formats: Level 1A (filenames start AST_L1A) is the rawest format and you must orthorectify these images to a DEM in order to calculate the sub-pixel offsets. Such orthorectification can be done in a variety of ways, but the instructions in COSI-CORR work well with the ENVI software. Level 14 (filenames start AST14DMO) includes images automatically orthorectified using the ASTER-derived DEM, which is of good quality for the Kashmir pair of images.

To download the data, ftp to asterweb.jpl.nasa.gov, login and passwd are aster. cd outgoing cd mike, cd pritchard. The data from the Kashmir earthquake is AST_L1A_00310272005055229_20080604154336_27202.hdf AST_L1A_00311142000060642_20080604154336_27203.hdf AST14DMO_00311142000060642_20080616134701_26452.tar.bz2 AST14DMO_00310272005055229_20080616134701_26454.tar.bz2 The data from Greenland are the other 4 files.

* Thanks to the authors of the Avouac et al., paper, you can directly compare your calculated offsets from the Pakistan earthquake with the results published in Avouac et al., 2006, which are available for download.

============================

HISTORY

This algorithms used to generate the the dense offsets using ROI_PAC were originally created by Mark Simons (Caltech) as a series of perl and matlab scripts. In 2009, Andrew Melkonian and Matt Pritchard (Cornell) added some capabilities to these scripts (including a georectification of the offset field, corrections for topography and capability to compute sub-pixel offsets of optical images) and converted the scripts to python. This is the first release of the software and so we expect that you may find problems with the software. Please report successes or failures with using the software to Andrew Melkonian (akm26@cornell.edu).

Offset Tracking (last edited 2009-09-04 15:07:54 by MattPritchard)