Processing Strategies

This page describes processing strategies and various options to set keywords in the "dot-proc" file.

Flowchart of the main processing steps in standard processing

The chart below shows the main steps of the process_2pass.pl perl script. Dash arrows indicate optional procedures.

.

New options for time series analysis ($do_sim and $do_mod)

This section describes a few options that will optimize the processing time when processing a series of pairs from the same track. When processing several pairs from the same track, only one simulation is needed and only one model (co-seismic or inter-seismic) projection is needed. You can now control these actions using the keywords $do_sim and $do_mod in the dot-proc file.

When processing the first pair of a series from the same track, use the default options:

do_sim = yes

do_mod = yes

The keyword $do_mod will be ignored if no model file is specified by keywords $CO_MODEL and $INTER_MODEL. If at least one model file is assigned, $do_mod = yes (default) will do the projection and scaling of the model data into a file and place this file in the simulation directory ($SimDir). Similarly, the keyword $do_sim = yes will tell the program to generate a new simulation file based on the DEM and place it in $SimDir.

For subsequent image pairs from the same track, you will set these keywords to "no":

do_sim = no

do_mod = no

and use the same $SimDir and $CO_MODEL and $INT_MODEL as for the first pair. This will indicate the program to use the existing simulation and model from the first pair, saving some cpu time and disk space.

If you choose to process all your pairs all the way to the geocoded product, all your interferograms will be ready to be used for stacking or in a time series analysis. If you choose to work in radar coordinates and stop the processing of each pair before the geocoding, all the unwrapped phase files will be in the geometry of their own 1st image. You will need to apply a small affine transformation to register them to the unique simulation file before stacking or using them in a time series. To do that, you may use the script rect_back.pl on each final unwrapped phase file. It will apply to the file the inverse of the affine transformation, which was defined by ampmag during the processing. All your "rect_backed" files will be co-registered to the unique simulation file and ready to be used in time series.

Note that if you want to geo-register a stacked phase or any deformation field in the simulation geometry, you need to generate a new geomap_Id.trans file with no affine transformation (Id stands for identity). Remember that the geomap.trans file generated when processing the first image pair is the transformation that goes from the 1st master image to the geographic coordinate system, not from the simulation to the geographic system. To create the geomap_Id.trans file, use make_geomap.pl with a .aff file set to the identity transformation (1 0 0 1 0 0).

Information about what baseline re-estimation does

Baseline re-estimation is done when you set "flattening=topo" in your dot-proc file. Some information is in the "ROI_pac Internals" document available on the ShortCourse page. The baseline re-estimation program "baseest" assumes an overall "zero" motion of the ground surface, and then it estimates a different baseline that best reduces the InSAR phase to near zero. Changing the baseline often results in something similar to a linear ramp being subtracted, but the pattern can be more complex than that with substantial quadratic or higher-order terms. You can see what phase was subtracted for the sim_ODR baseline in the "radar_ODR_*rlks.unw" file, and what was subtracted for the sim_SIM baseline in the "radar_SIM_*rlks.unw" file. Taking the difference of those two will show you the phase change that the baseline re-estimation removed. If you are trying to measure deformation that has a long wavelength compared to the size of your SAR images, then the baseline re-estimation will tend to remove the signal.

A new option for re-estimating the baseline with baseest using an incomplete a-priori deformation model ($unw_mod with $flattening)

When setting the keyword $flattening = topo, the program re-estimates the interferometric baseline based on the simulated topographic phase. This is a good option to remove orbit errors, especially when using data from satellites whose orbit is poorly known. However, if the phase includes a large spatial wavelength deformation signal, the baseline re-estimation will be biased. This option should be used in combination with an a priori model of the deformation (for example the SCEC velocity solution for Southern California), but you have to keep in mind that the baseline re-estimation will flatten the far field of your interferograms on the a-priori model. You will not learn anything new on far-field velocities using this option, but smaller spatial wavelengths features of the deformation field may still be of great interest. In some instances, you may want to mask out a part of the interferogram before running baseest to avoid additional bias due to improper a-priori model. For example, if you have a large earthquake in the middle of the scene, or if you know the far field plate motion but do not know how the deformation is distributed in particular fault zones, you may want to mask out these regions before running baseest. This can be simply done by applying a mask to the model to be removed from the phase. If you assign the value=0 to areas of the model where you do not know the deformation, these areas will be ignored in the baseline re-estimation. The keyword $unw_mod (default is yes) has been added to give the option of NOT removing the models from the phase before unwrapping when a model is used to re-estimate the baseline (set $unw_mod = no). This is necessary if the model does not cover the entire interferogram and a mask is used. For example, plate motion may be well determined in the far field of a fault and not near the fault, and the model mask (value=0) has to mask out the fault area. In such a case, the baseline is re-estimated using the total phase only where the model is not zero, but the unwrapping is done as if there was no model.

NOTE: $totalLunw.unw should always be the total phase. However, in the current version of process_2pass.pl it becomes the 'total phase minus the models' during the process and this is unfortunate.

How to remove deformation models

For more information on removing deformation models see this page Removing Deformation Models.

Manually picking offsets between SLC images

A common place for "process_2pass.pl" to fail is in matching the two SLC images to calculate the field of offsets to be used in "resamp_roi". See PickingOffsets for more information.

Processing Strategies (last edited 2012-11-13 22:16:38 by wildcard)