<< Click to Display Table of Contents >> Navigation:  Help Content > Interferometry Module > Interferometry Module - Frequently Asked Questions

Q. - The Interferometric Processing requires SLC products? Is the use of the Focusing module mandatory for the generation of these products?

 

A. - The Single Look Complex data to use as input in the SARscape Interferometric processing chain (e.g. Interferometry Module, Interferometric Stacking Module, ScanSAR Interferometry Module) can be either ordered as standard SLC product and then imported as such into SARscape or ordered as RAW product and then focussed using the dedicated SArscape module. It must be mentioned that not all original raw data are supported by SARscape.

 

Q. - Can I do Interferometry using any SAR acquisition pair where the two images cover the same geographical area?

 

A. - In general Reference and Secondary data must be acquired by the same sensor, in the same acquisition mode, with the same viewing geometry (same satellite track, same incidence angle) and with the same signal polarization; limitations related to baseline and doppler centroid difference must also be taken into account for a proper interferometric pair selection. Exception to these rules are:

- Data acquired by ERS-1, ERS-2 and ENVISAT can be used in the same pair. In case of ENVISAT-ERS/2 pairs, it is mandatory to execute the Interferogram Generation step by checking the flag "Coregistration with DEM"; it's worthwhile to mention that only pairs acquired during ESA dedicated ENVISAT-ERS/2 CInSAR tandem campaigns are suitable for interferometric processing.

- Two acquisitions with same polarization, coming from a single or multi-polarization data set, can be used in the same pair.

- ASAR Wide Swath and Image Mode acquisitions can be used in the same pair.

 

Q. - Can I do Interferometry using SAR data acquired by different sensors?

 

A. - In general you cannot form an interferogram between data obtained from different platforms, with different central frequencies and many other different parameters (e.g. orbits -> baseline, height; bandwidth etc.). The spectrum of the terrain reflectivity for a distributed target is generally completely uncorrelated from one frequency to any other one (even with just few Hertz apart), so for the Spectral Shift frequency you cannot mix data obtained with different central frequency.

 

An exception is combining ERS-2 and ASAR data. Here the two sensors have 30MHz of difference between the two central frequencies. In principle you could not form a coherent interferogram between two images from the two sensors (flying with 30min separation on the same orbit) since their bandwidth is around 16MHz (smaller than the center frequency difference), hence beyond the limits of the spectral shift. Nevertheless, with a baseline of around 2 km (ESA did dedicated campaigns for that) and knowing the spectral shift, it is possible to compensate for this difference and obtain interesting interferograms.

 

Q. - Is it possible to run Three or Four Pass Interferometry?

 

A. - Two, Three or Four Pass Interferometry are possible approaches to perform Differential Interferometric processing. What is implemented in SARscape is based on the use of a reference Digital Elevation Model (if this is not available, the use of the ellipsoid reference is also supported) for subtracting the topography (or simply the flat earth) from the initial interferogram (_int). Depending on how the reference DEM has been generated (i.e. InSAR techniques or other sources) the program executes the 2-, 3-, or 4- pass Differential Interferometry approach.

The removal of the phase component due to the topography is needed in order to "isolate" the phase component due to the displacement.

 

Q. - How can I properly select a data set for the Dual Pair Differential Interferometry processing and how is it implemented?

 

A. - This module allows to provide as input two Secondarys and one or two References (i.e. 3- or 4-pass approach respectively); the functionality allows to automatically perform all the steps from the interferogram generation to the phase unwrapping for the two pairs and, at the end, the program generates at once the Digital Elevation Model and of the Displacement Map from the two unwrapped interferograms by inverting a simple linear system (the concept is similar to that implemented in the SBAS using a more robust approach, since the SBAS uses an over-determined system with many measures to make the estimation more accurate and reliable); all products are finally geocoded.

Concerning the "ideal" input dataset, it is represented by: i) a pair good for DEM generation (e.g. temporal separation as short as possible to get high coherence, no displacement between the two dates and baseline not too small to have good height sensitivity), which we can call the "DEM pair"; ii) a pair good for displacement detection (e.g. a co-seismic pair, with one acquisition before and one after an earthquake and possibly a quite small baseline), which we can call the "DInSAR pair". The functionality allows to select different possible models depending on the displacement type/dynamics: for instance the "linear model" is suitable to describe subsidence phenomena; the "step model" for acquisitions spanning an earthquake event; the "no model" configuration, where the first pair (used for the DEM generation) did not record any displacement which was vice versa captured by the second pair (used for the Displacement Map generation).

 

Q. - Is it possible to process data with the same polarization acquired in Single and Multi-polarization Mode?

 

A. - It is possible to combine, in the same interferometric pair, a single polarization with a multi-polarization acquisition; for instance, in case of ALOS PALSAR data, we can make a pair using the HH channel of a Fine Beam Single (FBS) and the HH channel of a Fine Beam Dual (FBD).

 

Q. - When working with PALSAR FBD InSAR pairs, what is the best polarization to use for DEM generation?

 

A. - In general, independently from the acquisition wavelength, it is better to use one of the two co-polarisations (HH or VV) and, between the two, the HH has usually a stronger backscatter and a better coherence.

 

Q. - What is the Accuracy/Precision achievable using InSAR techniques for Digital Elevation Modeling?

 

A. - The accuracy (in terms of interferometric phase reliability/quality) depends essentially on the coherence of the InSAR data pair. The accuracy (intended as DEM precision/resolution) will be driven by several factors, among which the most important are:

❖ Pixel spacing (the smaller the better precision in x and y direction).

❖ Height of ambiguity (the smaller the better precision in z direction), which depends mostly on acquisition baseline and wavelength. A reasonably good result is to generate a DEM with a precision (in the height estimate) between 1/10 and 1/20 of the 2π ambiguity.

 

Q. - The height figures provided with the final Digital Elevation Model are Absolute or Relative Values?

 

A. - They are absolute elevation values. It has to be pointed out that the final output DEM will be referred to the ellipsoid if the reference (i.e. input DEM), was provided with ellipsoidal heights (geoid subtracted). Nonetheless a specific tool is available to add the geoid component to an ellipsoidal DEM.

 

Q. - The Baseline values which are estimated with different versions of the software are slightly different each other and also different from those which are reported in the data provider catalogue. Why this happens and does this influence the interferometric processing accuracy?

 

A. - The normal baseline is an approximate and indicative value (coming from a linearization performed around a certain reference point, e.g. center scene with given reference height); that allows to get good general information on a certain interferometric pair, but it is never used during the "real processing" (e.g. interferogram generation and flattening, phase2heigh conversion, PS/SBAS processing etc.) within SARscape, in order not to introduce systematic errors due to the approximation; in the real processing the original orbits from the state vectors and their variation along the full frame, their non-parallelism and many other parameters must be carefully considered. When the geometry is concerned, during the SARscape processing the original Range-Doppler equations of the different acquisitions are used and jointly solved, without introducing any simplification (like the normal baseline).

 

Q. - How can I choose an Appropriate Baseline for doing InSAR or DInSAR processing?

 

A. - The criteria to select a SAR pair with an "optimal baseline" (we refer here to the perpendicular or normal baseline) differ depending on the objective of the interferometric processing.

When doing InSAR processing (i.e. Digital Elevation Model Generation) we would like to have a large baseline; in essence the larger the baseline the better the capability to detect small height changes (refer to the Ambiguity Height formula below). Theoretically speaking the maximum baseline limit, which can never be exceeded, is the critical baseline (refer to the Critical Baseline formula below); practically speaking one should avoid using baseline larger than half the critical value.

 

Bcr = λ R tan(θ)                                     Hamb = λ R sin(θ)    

 2 Rr                                                        2 Bn

 

Where Bcr is the Critical Baseline; Bn is the Normal Baseline; λ is the acquisition wavelength; R is the slant range distance; Rr  is the pixel spacing in slant range; θ is the acquisition incidence angle; Hamb is the Ambiguity Height. Of course the more the land cover/topography conditions are critical (i.e. dense vegetation/steep slopes) for the interferometric technique application, the more problems shall be introduced by using an InSAR pair with large baseline...

 

When doing DInSAR processing (i.e. Displacement Mapping) we would like to get a very small baseline (best being a 0 meter baseline) in order not to have topography induced fringes in the interferogram.

 

Information related to Critical Baseline, Normal Baseline and Ambiguity Height of a SAR pair can be retrieved by using the Baseline Estimation functionality.

 

Q. - What is the meaning of the 2π Ambiguity Height and how is this value computed?

 

A. - It represents the height value, which provides a phase variation of 2π (one interferometric cycle). It is inversely proportional to the baseline and it depends on the wavelength, incidence angle and other acquisition parameters. If we are observing a topographic phase, which means that we have flattened our interferogram using only a flat earth (without DEM), our interferometric fringes resemble contour lines separated each other by a 2π ambiguity distance.

 

Q. - What is the meaning of Interferometric Cycle?

 

A. - Observing a black and white flattened interferogram, we see all the iso-tone lines (“fringes”) that correspond to the wrapping of the phase. If we consider a mountain side, and we assume we move (“climb”) from one black phase pixel through all the gray levels up to white phase and then back down to black, we move from one phase value to another that is 2π radians (or one interferometric cycle) larger than the first one.

 

Q. - What is the meaning of Critical Baseline?

 

A. - The critical baseline is the theoretical maximum value above which, specifically for the input data set, distributed targets are not correlated anymore (i.e. coherence=0).

 

Q. - If I have a Data Temporal Series (i.e. 20 acquisitions), which I want to process in interferometric mode, but I'm interested only on a small portion of the full frame; is it possible to define the area of interest in one scene/acquisition and get the other 19 images automatically cut on the same area?

 

A. - This can be done using the Sample Selection from the Interferometry module tools. Using this functionality the area of interest is defined in the first acquisition of the input list, then the program will coregister the other input data and will execute the subset on the same geographical area.

 

Q. - In case I have to perform the PRF Correction, when should it be done: on SLC data before the Interferogram Generation step, or rather on coregistered data after the Interferogram Generation step?

 

A. - We found that in some cases geocoding Radarsat-1 data with 1 GCP produced an accurate geolocation close to the GCP while the product became shifted if observed in areas far from the GCP (this was evident when moving, from the GCP position, in azimuth direction). We attributed this problem to a wrong value reported for the Pulse Repetition Frequency (PRF) and, for this reason, we introduced the PRF Correction tool. Considered that, we would suggest to check - running a Geocoding process - whether the above mentioned geolocation problem exists using one GCP only; if it does not, you do not need to correct the PRF. Vice versa if you identify a geocoded product shift in areas far from the GCP it means that you must correct the PRF. This correction has to be executed before running the Interferogram Generation step.

 

Q. - I am trying to make an interferogram in an area of High Relief and I am checking the "Coregistration with DEM” flag in the Interferogram Generation . The output _dint file still shows a Phase Ramp; is it possible to get rid of it?

 

A. - The main purpose of the "Coregistration with DEM" option is to improve the Secondary to Reference coregistration in areas of strong relief; this is especially effective when processing very high resolution data (e.g. COSMO-SkyMed, TerraSAR-X) or working with long orbit segments or working with ENVISAT-ERS/2 pairs or processing data acquired at high latitudes or using non zero-Doppler annotated data (especially with long wavelength). Nevertheless, in case you have a phase ramp due to orbital inaccuracies, it will not be removed unless you execute the Refinement and Re-flattening step (after the "Phase Unwrapping"). It is possible to automatically remove the residual phase, which possibly still exists after the removal of the phase component due to topography and/or flat earth (_dint and _fint files), by checking the "Remove Residual Phase" flag in the Preferences>Flattening panel or by executing the Remove Residual Phase Tool on the differential interferogram (_dint or _fint).

 

Q. - Do I need to Coregister Reference and Secondary Data of my interferometric pair, using the relevant Basic module functionality, before initiating the interferometric processing?

 

A. - You do not need to run the Basic module Coregistration, since the Secondary to Reference coregistration is performed during the "Interferogram Generation" step.

 

Q. - How can I check the Coregistration Accuracy of the Secondary to Reference interferometric data?

 

A. - A first visual check can be done by displaying - in two linked viewers - the Reference and Secondary coregistered images (Multilooked or Single Look Complex products), which are generated as output of the"Interferogram Generation" step.

A more in depth analysis can be carried out by analysing the output shape file (_winCoh_off.shp), which is generated as output of the "Interferogram Generation" step. This provides information related to the signal/noise ratio and coherence values relevant to each coregistration window, as well as the shift calculation in range and azimuth direction.

 

Q. - In the Interferogram Generation step, the program ends with an error message?

 

A. - The error message reported during the "Interferogram Generation" step, is typically due to a Secondary-to-Reference coregistration problem. This can be related to several reasons, which can be summarised as follows:

 

1.Large orbital inaccuracies, which cause a wrong shift calculation during the coregistration initialisation. In order to overcome this problem just open the Preferences>Coregistration panel and deselect the "Initialisation from Orbit" flag. In case the coregistration should fail again, just increase the number of coregistration windows in azimuth and range direction (upper part of the "Set Default value>Coregistration" panel) up to respectively 30 and 20 or more.

2.The Cross Correlation Central Window is not large enough (in azimuth and/or range direction) for the shift initialisation; this case can be reported when the "Initialisation from Orbit" flag is not selected. Note that the Cross Correlation Central Window must be at least two times bigger (in azimuth and range direction) than the distance between the same pixel in the Reference and Secondary not coregistered data.

3.Large portions of the scene (typically homogeneous areas such as water, forest, sand, etc.) lack of spatial features, which are required for calculating the cross-correlation function between Reference and Secondary file. In these cases it is possible to manually locate points (Coregistration file), representing the center of the coregistration windows, in those areas where cross-correlation features (e.g. scatterers such as rocks, urban settlements and other man made objects) exist.

4.A large difference in the doppler centroid of Reference and Secondary acquisitions. This kind of problem is sometime reported using ERS-2 data acquired after the year 2002, when the satellite started being operated without any gyroscope available (reduced satellite orbit stability). In order to check what the doppler centroid difference is, just execute the  the "Baseline Estimation" step: in case the Doppler centroid difference is higher than the critical one (i.e. value of the Pulse Repetition Frequency reported in the data header file - .sml) it means that the SAR pair is not suitable for interferometric processing.  

5.A large temporal de-correlation between Reference and Secondary acquisitions. This is typically reported when most of the imaged area is made of features, such as densely vegetated areas or water, which are subject to changes if observed at a certain time distance (i.e. SAR interferometry multi-pass configuration); temporal de-correlation related problems tend to become more important when Reference and Secondary acquisitions are separated by very long time distances (i.e years). Generally the temporal de-correlation decreases by increasing the acquisition wavelength (e.g. from C band to L band).

6.The number of coregistration windows in azimuth and range direction, which are used for the fine shift estimate (Preferences>Coregistration>Fine Shift Parameters section), is not sufficient; this case is typical of poorly correlated data (see point n. 5). Increasing this value (e.g. 40 windows or more) can solve the problem; however it must be mentioned that poorly correlated data cannot provide reliable results in terms of interferometric phase.

7.Very large normal baseline (i.e. half of the critical value or more). The execution of the "baseline estimation" step allows calculating this value and comparing it with the critical one.

 

Q. - The Flattened Interferogram (_dint) seems to only have the "curved Earth" phase removed, not the topographic phase. Is that correct? Does SARscape produce a differential interferogram where also the topographic phase is removed?

 

A. - The differential interferogramcan be generated using an input Digital Elevation Model or the ellipsoid height (it can be 0 or else, depending on the mean altitude of the imaged area). In both cases the program removes, from the original unflattened interferogram (_int), the flat earth phase and, only when the DEM is provided in input, it also removes the phase component due to the known topography.

It must be mentioned that the topographic phase removal depends on the quality and accuracy of the input Digital Elevation Model as well as on the correspondance between the geolocation of DEM and SAR Reference image (see also the following answer).  

 

Q. - Why the differential interferogram (_dint), shows up plenty of Topographic Fringes? Should not have they been separated as synthetic phase (_sint)?  

 

A. - During the flattening process the reference input DEM is re-projected to the SAR Reference acquisition geometry (slant range). In case this re-projection is not accurate, due for instance to inaccuracies in the orbital parameters, the slant range DEM will not fit with the SAR geometry; this introduces "flattening errors" (i.e. under-flattening and over-flattening fringes). For this reason the use of precise orbits is strongly recommended. In case precise orbits where not available and the standard ones were not enough accurate, then the use of a Ground Control Point is required to successfully carry out the flattening process. It is important to note that the GCP is not needed if the manual or the automatic correction procedure has been previously executed.

 

Q. - Why the differential interferogram (_dint) shows several large fringes, resembling a Phase Ramp, regularly distributed throughout the image?  

 

A. - This effect is due to orbital inaccuracies. The phase ramp (i.e. orbital fringes) is not present if precise orbits are used. Residual phase ramps can be removed by carrying out the Refinement and Re-flattening step.

 

Q. - What the Remove Residual Phase Frequency step consists of?  

 

A. - This step can be carried out either automatically (during the flattening process), whether the relevant flag in the Preferences is checked, or manually (by means of the proper Interferometry Tool) using the _dint or _fint file as input. It is aimed at removing those residual fringes (phase ramp), which remain after the flattening process. The peak in the frequency domain (both in range and azimuth direction) is estimated by using a Fast Fourier Transform on each window whose dimensions are provided in the SARscape panel; afterwards a fitting is performed on the frequency values computed for each window in order to calculate the phase ramp to remove from the whole image.

 

Q. - What is the Fake GCP?  

 

A. - This is a Ground Control Point, which is automatically generated by the software, during the interferogram flattening process, in order to correct the azimuth start time and the slant range distance of the Secondary image. It does not change any parameter of the Reference image.

This automatic procedure is activated, in the Interferometry Module, when the Automatic Secondary Orbit Correction flag is checked in the relevant Preferences; this is valid also for the Persistent Scatterers processing. The automatic correction is always performed in the SBAS processing (i.e. independently from the Preferences setting).

 

Q. - Is there a way to find the SNR (System noise or Temperature noise) value, Residual Topographic Phase Noise value and Processing Noise value of an Interferogram?  

 

A. - SARscape does not discriminate among the different types of noise. The SNR value is related to the interferometric coherence (γ) on the basis of the following formula:  SNR = γ2/1-γ2. The coherence value, which is reported in the "..._cc" output of the "Adaptive Filter and Coherence Generation" step, can vary from 0 to 1; this value is inversely proportional to the systemic spatial de-correlation (i.e. the additive noise) and to the temporal de-correlation between Reference and Secondary acquisitions.

 

Q. - In the Adaptive Filter and Coherence Generation step, which of the three proposed approaches (i.e.  Boxcar or Adaptive window and Goldstein) has to be preferred and when one method can perform better than the others?

 

A. - In most of the cases the best choice is for the Boxcar (very high coherence pairs) or the Goldstein. The Adaptive approach can be adopted in case one wants to adapt the filtering window to take into account for the stability (or stationarity) of the backscatter value (i.e. SAR amplitude); this method, which typically requires several trials in order to find the optimal processing parameters, can provide better results with very high resolution data (e.g. TerraSAR-X or COSMO-SkyMed).

The use of the Goldstein filter is recommended especially in those cases where the fringe pattern is hard to detect due to temporal or baseline related decorrelation or to challenging land cover or morphology conditions. This approach is often preferred for its adaptability to different coherence conditions.

 

Q. - I am using the Goldstein filter during the Adaptive Filter and Coherence Generation step. I noted that, in the resulting coherence image, the value of every pixel is 1. Do you have any suggestion about what might have caused this strange coherence behavior?

 

A. - Whether the filtering action is quite strong, like in the case of the Goldstein filter, and the processing parameters are set in order to dramatically reduce the phase noise, most of the pixels in the resulting coherence image can be saturated to 1. To avoid this, the coherence can be generated by:

1.Setting the Goldstein filter parameters in order to reduce the filtering strength.

2.Using the Boxcar approach.

3.Uncheck the "Coherence from Fint" flag in order to get the coherence generated from the unfiltered interferogram (dint).

 

The newly generated coherence can be used in association with the previously generated filtered interferogram.

 

Q. - I noted anomalous "square-box" like features in the interferogram, which was filtered using the Goldstein method. What can be the reason and how can be these artifacts possibly avoided?

 

A. - It can happen that, depending also on the quality of the input interferogram, a strong filtering setting can introduce such artifacts. These are possibly avoided (or reduced) by increasing the "Windows Overlap Percentage" parameter.

 

Q. - Reading the reference literature available for the Goldstein filter, it appears that the alpha value typically lies in the range of [0, 1]. Why the default range of alpha variability (Preferences>Adaptive Filter>Goldstein Interferogram Filtering) considers values which can be much higher than 1?

 

A. - The alpha value can possibly exceed the typical range of [0, 1]. This happens where the coherence is very low and the interferogram is eventually very noisy. As a matter of fact the SARscape default alpha max, which is used where the coherence is 0, is higher than 1 (around 2.5 or more); nevertheless the value which is adopted in coherent areas is always lower than 1, as it varies linearly from the alpha min (high coherence zones) to the alpha max (not coherent zones).

 

Q. - How the interferometric fringes can be displayed in colour?

 

A. - A standard RGB colour table, which is typically used to represent the interferometric fringes, is applied when transforming the complex interferogram (_int, _dint or _fint) into an output 8 bit Tiff image using the relevant SARscape Tool.

 

Q. - Is it possible to Mosaic Interferograms generated from InSAR pairs of overlapping scenes?

 

A. - The Mosaic tool works for geocoded real data. It means that, in order interferograms to be mosaiced, the following two processing steps must be previously carried out:

❖ Geocode the complex interferogram.

❖ Split the complex interferogram into phase and module components.

The phase and the module components have to be mosaiced separately; afterward the mosaiced phase and the mosaiced module shall be combined again to form a complex interferogram.

 

Q. - Is it possible to make a Difference between two Interferograms?

 

A. - The Interferogram Difference can be used for this purpose.

 

Q. - We see that the Coherence is influenced also by topography and backscatter intensity. Is there a way to get rid of these influences?

 

A. - One possibility is to check Spectral Shift Filter flag, which is the default SARscape setting. Moreover it is possible to modify the default parameters relevant to the Local Frequency Removal

 

Q. - A Coherence Threshold can be set in different step of the processing chain. Is it better to apply always the same threshold value or it is better to apply different thresholds?

 

A. - In general it is better to adopt different coherence thresholds depending on the specific processing steps, in particular:

❖ In the Phase Unwrapping a low threshold (e.g. around 0.2) shall be preferred in order to leave more freedom to the region growing algorithm to diffuse without blockages (i.e. low coherence areas). This will reduce the possibility to have "phase jumps", which should be edited afterward.

❖ In the Phase to Height Conversion and Geocoding as well as in the Phase to Displacement Conversion and Geocoding, a higher threshold can be set in order to mask - in the final interferometric product - those areas where the coherence is low and consequently the phase measurement is less reliable; thus the specific value here depends on the reliability one wants to associate with the DEM or the Displacement Map. Value of 0.3 or more generally provide reliable results.

 

Q. - The Phase Unwrapping step failed due to a memory allocation error. Why this happens and how can the problem be solved?

 

A. - The error is due to a limitation of the WINDOWS operative system in handling the unwrapping operation of large data (in terms of file size). This problem can be overcame by multilooking and undersampling the data, using the "Decomposition Levels" option.

 

Q. - Does the Minimum Cost Flow Unwrapping method implement the SNAPHU algorithm or something similar?

 

A. - The Minimum Cost Flow algorithm implemented in SARscape is derived from the published work of Mario Costantini (i.e. "A novel phase unwrapping method based on network programming"). Unlike the SNAPHU program, which works by decomposing the image in tiles without modifying the original pixel spacing, the SARscape implementation foresees the use of the Decompositions Levels. This implementation enables to iteratively decrease the resolution (i.e. a factor 3 multilook and under sample for each decomposition level) in order to both enable the processing of large data and homogenize the unwrapping process of the entire imaged area. The original resolution is reconstructed by iteratively recovering the information from the previously multilooked and under sampled "Decompositions Levels". In case the phase image to unwrap is characterized by very frequent displacement variation, the "Decomposition Level" option must be handled with care; indeed in this case the displacement variations can be aliased, and this happens especially when the number of decompositions is set higher than 2.

It has to be pointed out that the "Decomposition Levels" approach can be adopted either with the Region Growing or with the Minimum Cost Flow method.

The use of the Delaunay triangulation method can be considered in case of diffused low coherence areas.

 

Q. - Is there an easy way to create the Orbital GCP File, which is used as input in the Refinement and Re-flattening step? What is the process, which the program performs during this step?

 

A. - The most important concept in the identification of the ground control points collected for the "Orbital GCP file" it is to locate them in areas where there is no evidence of residual topographic fringes and far from eventual displacement areas; for this reason one of the SAR interferometric products useful for this purpose it is the flattened (and possibly filtered) interferogram (_dint or _fint). Moreover the points have not to be located over "phase islands" since the program works on the unwrapped interferogram and thus erroneous phase jumps are erroneously interpreted during this step; for this reason another SAR interferometric product to check during the point collection process it is the unwrapped phase (_upha). Finally it is important to spread as much as possible the points throughout the imaged area.

The points can be found in the SAR image geometry, without entering any displacement or height value; in such case the program will consider zero displacement and the height value written in the reference DEM. The points can be also entered in cartographic co-ordinates, this is for instance an option when the same points have to be adopted for different interferometric pairs located in the same geographical area (e.g. an ascending pair and a descending pair); in such case the program will automatically re-project, in the slant SAR viewing geometry, the range and azimuth position of each point.

It is possible to choose between two methods for the re-flattening process: i) the orbit correction (more precise since it relies on the satellite orbital position); ii) the residual phase removal (more robust, but also more rough since it can only remove a phase ramp or a constant phase offset depending on the polynomial degree adopted). The default process adopted by the program it is to attempt by the first method and, if the orbit correction it is not sufficiently precise or the two orbits are too close each other (i.e. small normal baseline, which means the impossibility to find "realistic" solutions) switch automatically to the second method. However it is possible to force the program adopting one or the other method as well as to modify the polynomial degree or the automatic checks, which the program performs to estimate the GCP location accuracy; this can be done by changing the default setting in the Input Parameters>Refinement and Re-flattening section.

The minimum required number of points actually depends on both the Refinement Method , which have been set in the relevant Preferences: when the "Residual Phase" refinement method is checked, the number of points must be at least equal to the "Residual Phase Poly Degree"; when the "Orbital" refinement method is checked, the number of points must be at least equal to 7; when the "Default" refinement method is checked, the software automatically set the refinement method depending also on the number of points. In any case a number of points higher than the minimum is recommended, in such case the solution will be an averaged and more reliable correction.

 

Q. - What is the reason why I observe Phase Islands after the unwrapping process and it is possible to avoid them?

 

A. - Actually, either inside or outside these "islands", the unwrapping is "locally" well performed. However it remains a discontinuity (i.e. a finite number of 2π phase cycles) on the island border. This kind of problems typically appears when the phase is noisy along the border, or when the interferometric fringes are very dense (i.e. the phase change is very fast) and thus the correct phase jump is wrongly estimated. It must be also noted that in some instances the jumps are correct since there is a real phase discontinuity. The use of the Delaunay unwrapping method is generally the best choice for limiting these phase artifacts.  

 

Q. - How is it possible to check the quality of the Refinement and Re-flattening step?

 

A. - Once this processing step is completed the Root Mean Square error (BFRMSerror), which is calculated from the difference between the height value of the Ground Control Points and the corresponding value in the interferometric phase, is provided. Root Mean Square errors ranging from around 2 to around 10 are a good preliminary indication that the Refinement and Re-flattening was successfully executed (i.e. good GCPs). A further possibility to visually check the quality of the result, it is to inspect the re-flattened interferogram in order to see if the orbital related fringes have been removed from the original flattened interferogram (i.e. _dint or _fint file).

The orbital refinement step (that has in any case to be performed, also in case of precise orbits, to estimate the absolute phase offset) produces corrected orbits which are not necessarily the one closest to the true ones; the "refined" orbits are those which describe the phase data at the best, compared with the input DEM.

It must be noted that the "Refinement and Re-flattening" using pairs with small baseline values (i.e. less than around 50 meters) is critical.

 

Q. - In the Refinement and Re-flattening step, is it preferable to manually enter the co-ordinates and height of Ground Control Points accurately collected on the ground (or onto a topographic map) or to use the reference Digital Elevation Model previously transformed in the Reference image slant range geometry?

 

A. - During this processing step we compare the height obtained from the SAR phase with the reference one from the GCP (and/or from the reference DEM). We know that, even with very good coherence, and with quite large baselines, we will never get from the SAR interferograms a height with an accuracy of 1 cm, but say of few meters. This means that, if we compare a GCP height with a precision of 1 cm with a “phase height” with a final expectable accuracy of few meters, we see that there is not reason to look for the "super-accuracy" of our GCPs. In other words, in most of the cases, the result of the processing with a good reference DEM (e.g. often already SRTM-3) is good enough.

It must also be considered that, once the orbital refinement is applied, we do not estimate just phase 2D polynomials to subtract (as estimation of the residual phase from the orbits), but we use an orbital model to correct the inaccuracies. This means that we do not apply all possible corrections, but only those in agreement with a physically consistent orbital configuration.

 

Q. - I have compared a DEM generated with SARscape against a validated high accuracy DEM. I observed a strange and may be Systematic Difference  between their height values. Can you explain why?

 

A. - There are two possible arguments to explain the discrepancies between the height in your SARscape product with respect to that reported in the high accuracy reference DEM:

1.It is related to the GCP location in the "Refinement and Re-Flattening" step. It is worthwhile to recall that, once the "Refinement and Re-Flattening" processing step is completed, you can do a first "quality assessment" by means of the: a) "Root Mean Square error", which is calculated from the difference between the height value of the Ground Control Points and the corresponding value in the interferometric phase (an error ranging from around 2 to around 10 is a good preliminary indication that the GCPs have been properly selected); b) "Absolute Phase Offset", which represents the difference between the interferometric phase value and the fitted value based on GCPs; c) "_refinement.shp" file, which contains additional and useful information for the assessment of the correction parameters calculated from the input GCPs (refer to the online help for more details).

2.It is related to the reference surface of your SARscape product. In the Interferogram Flattening process, where the know topographic phase component (_sint) is separated from the unknown/differential phase (_dint), the input DEM is normally referred to the ellipsoid. This means that the final InSAR DEM will be referred to the ellipsoid and this could explain the systematic difference with respect to your reference precise height, which is probably referred to the geoid. It is worthwhile to mention that it is possible to add the geoid component to a SARscape DEM.

 

Q. - If I generate two times a DEM, with exactly the same data and same processing configuration, but with a different set of GCP points in the Refinement and Re-flattening step, the output DEMs could have notably different elevation values (up to 20 meter differences have been found). Are such differences expected and is it possible to prevent them?

 

A. - The Refinement and Re-flattening can be a really "delicate and sensitive" tool... What happened in your case is that, choosing a different set of GCPs, a different phase offset has been estimated and a different average height has been eventually associated to the two DEMs that you've generated.

Therefore, much attention must be paid when the GCPs are selected (the most important selection criteria are described in the relevant section of this document). A possible reason of the differences that you measured, it can be due to the fact that one or more GCPs have been located in areas where there is some residual topography in the differential phase (i.e. _dint, _fint or _upha).

It is worthwhile to recall that the input GCPs can be entered either in SAR geometry or in any supported cartographic reference system. In this second case the program automatically takes care to convert the co-ordinates of each point into the reference slant range reference geometry.

 

Q. - How is it possible to check and verify the Absolute Phase Offset (in radians), which is calculated in the Refinement and Re-flattening step?

 

A. - One of the output products, which are generated in the Refinement and Re-flattening step, is the _refinement.shp. It provides information useful for the assessment of the correction parameters calculated from the input GCPs. In particular the "AbsPhDiff"  and the "PhaseDiff" represent respectively the absolute and the relative difference (in radians) between the real phase and its fitted value based on GCPs.

 

Q. - The fringe pattern changes dramatically when going from the _int (original "raw" interferogram) to the _dint (flattened interferogram) through the _upha (unwrapped phase) and all the way to the final DEM; what is the Overall Processing Approach and the philosophy behind it?

 

A. - The approach adopted in SARscape it is not aimed at a linear combination of Flat-Earth + DEM phase; the objective is to go back to the full interferometric (absolute) phase (adding the _sint , the _upha and the absolute phase offset) in order to exploit the original physics of the acquisitions (i.e. the combined Reference and Secondary, Range and Doppler equations). In this perspective the InSAR DEM is basically equivalent (and precise) when exploiting a reference DEM for the flattening or just a constant reference height.

 

Q. - What is the approach used in the Wavelet Combination DEM?  

 

A. - The standard wavelet approach is applied with the assumption that, typically in an interferometric high resolution DEM, the atmospheric artifacts (as well as possible height offsets and residual ramps) affect mostly the low frequencies. These artifacts can be removed (or reduced) by means of a reference low resolution DEM (e.g. GTOPO30 or others), which is used to correct the low frequency of the high resolution product while keeping untouched the spatial detail coming from the high frequencies.

 

Q. - I want to generate an InSAR DEM over a tropical (densely vegetated) area in Cameron; the objective is to detect the Coastline. I plan to use ALOS PALSAR FBS data. Do you have any helpful suggestion?  

 

A. - The main problem is definitely related to the very dense vegetation, which most probably characterize your area, thus I would suggest minimizing the temporal de-correlation by choosing the minimum possible time interval (46 days) for your PALSAR pair. Moreover, if there is the chance, do select an acquisition period when the rains are not to heavy and persistent. It is finally worthwhile to mention that a good help in detecting (and characterising in terms of land cover) the coastline can also be provided by the use of amplitudes and coherence data.

 

Q. - I want to generate an InSAR DEM over Iceland. Do you have any special advice to provide?  

 

A. - In general the usual criteria apply also in this case: i) the temporal distance between the two acquisitions should be as short as possible; ii) data should be acquired in a season of the year when atmospheric conditions are stable and major meteorological perturbations (e.g. storms, snow falls, etc.) are not expected. Moreover, in this special case (high latitudes, i.e. Iceland), an important issue is related to the ice motion; indeed some fringe patterns over glaciers can be misinterpreted as height while being related to movement.

 

Q. - I’m implementing the high resolution TerraSAR-X images to get InSAR DEM for the Estimate of Building Heights, but I'm not happy with the output product accuracy; can you provide any suggestion on this topic?  

 

A. - The use of SAR data for building height estimate in urban areas is not the most suitable approach, especially due to the continuous alteration of layover and shadow areas, which are due to the buildings themselves; moreover the unwrapped phase typically requires extended manual corrections (phase editing).

 

Q. - How to Combine Two Geocoded Displacement Maps in order to get a single output product?

 

A. - If the two maps are partially overlapping while being acquired on two different locations, it is possible to mosaic them by checking the option "Precision" in the SARscape>Tools>Mosaicing>Conventional Mosaicing functionality. In this way the two maps are combined by weighting the pixels on the basis of the "..._precision" file, which is generated among the outputs of the Phase to Displacement Conversion and Geocoding step.

The _precision output is derived from parameters such as coherence and wavelength; it provides an estimate (i.e. standard deviation value) of the measurement precision. The higher this value the lower the measurement precision.

If the two maps are fully overlapping (same location/coverage of the imaged area) and the objective is get a single better map, then the option "Mean" in the SARscape>Tools>Mosaicing>Conventional Mosaicing functionality can be used. This approach can be exploited for instance in order to fill some low coherence areas of one map with data that might be present in the other one, while averaging the areas covered by both maps and then reducing some noise there.

 

Q. - What are the criteria (and the limitations) to convert the Line Of Sight Displacement Into A Displacement On A User Defined Direction?

 

A. - SAR is measuring the LOS (Line of Sight) displacement, that means the component of the full displacement D (a 3D vector) projected onto the LOS direction. Assuming a certain angle (we can call it β) between the LOS and the D direction, we measure a displacement |LOS| = |D| * cosβ, which represents the scalar product between the LOS and D vector. This means that, knowing the direction of the D vector and using the acquisition geometry to get the direction of the LOS, we can estimate the original magnitude of the full displacement as |D| = |LOS| / cosβ

As an example, if we have an area affected by subsidence phenomena and we want to estimate its magnitude, we can assume that the direction of the D vector is in this case vertical, and we can easily solve the conversion problem.

Vice versa, if the real displacement did not occur along the vertical direction (or more in general along an a-priori well known direction), there are infinite vectors (one for each possible value of β) that projected onto the LOS give the value we measure; each of them has a different original magnitude and each of them has a different component when projected onto the user defined direction; this means that in this case, and using only one acquisition geometry, we cannot assess the real displacement.

In other words, it makes sense to use this option only if the real displacement on the ground occurred along a direction that can be accurately determined; otherwise not only the re-projected displacement direction, but also its magnitude will be wrongly estimated.

 

Q. - Is it possible to Merge the Intensity Data with the Interferometric Fringes in order to visualise them in the same image?

 

A. - RGB combinations showing the interferometric fringes draped onto the SAR amplitude features can be obtained by following the steps below:

1.Separate the filtered complex interferogram (_fint) into phase and module components (Tools>Conversion - Complex to Phase and Module)

2.Use the amplitude (Reference or Secondary image) as module for regenerating a complex datum with the previously separated interferometric phase (Tools>Conversion - Phase and Module to Complex); note the amplitude image can be previously despeckled (_pwr_fil) for a better visualisation.

3.Generate the Tiff file (Tools>Generate Tiff) by properly setting the "scale" and "exponent" factors (values of respectively 0.3 and 0.5 are typically suitable for this purpose): an optimal tuning of the histogram stretching enables to have the best visualization of both the interferometric fringes and the amplitude image texture. Note that an enhanced visibility of the amplitude features is usually achieved by reducing the "scale" or increasing the "exponent".

 

Q. - Is it possible to calculate the Velocity of a Large River using SAR interferometry?

 

A. - If the river is not frozen, any repeat-pass interferometric pair (with temporal distance larger than a few seconds between the two acquisitions) will have not coherence over the water, so none possibility of measuring any meaningful phase with any interferometric technique. The only option to measure such effects is with “quasi-simultaneous”, “along-track” acquisitions, like those that can be obtained (from satellite) with the experimental split-antenna modes of TerraSAR-X and Radarsat-2, and in some of the orbits from the TanDEM-X constellation. Having such data (or airborne data with a specific along-track baseline configuration between two antennas) one can directly use the available SARscape modules for InSAR and DInSAR processing to obtain such measure.

 

Q. - Is it possible to Monitor Subsidence phenomena (also on the basis of historical data), whose rate reached peak values of few meters per year. The area of interest, which is exploited for coal mining extraction since about 20 years, is partially covered by dense vegetation and forest. Could you also give a rough Estimation of the Accuracy that we can achieve?

 

A. - You can definitely exploit different kind of SAR interferometric techniques (from the classical two dates approach to the more advanced Persistent Scatterers and Small Baseline methods) for your purposes.

One of the major factors, which drive the displacement measurement accuracy that you can reach with SAR Interferometry, is the observation wavelength; indeed you can apply this technique on SAR data acquired in X-band, C-band or L-band data. In general we can say the displacement measurement accuracy is in the order of the cm for the classical two dates interferometry and in the order of the mm for the Interferometric Stacking (PS or SBAS); note that, in order to apply the Interferometric Stacking methods (mm accuracy), you have to rely on an interferometric temporal series and thus you first need to check how many data acquired in interferometric mode are available on your area of interest.

Therefore you have to check the archive data (e.g. ENVISAT ASAR, ALOS PALSAR or other satellites) availability over your area of interest. These archive data are essential to build a kind of reference baseline map of the subsidence distribution in the past years: typically, in coal mining area, the subsidence follows - with some temporal delay - the underground extraction direction.

Afterwards you can think about future acquisition planning to monitor the current subsidence distribution and rate. In my opinion, before deciding on topics such as the most suitable satellite/wavelength, the best temporal frequency, etc., it is better to analyse the past subsidence distribution and detection with the available SAR archive data (see previous point); this will also give you a better and more realistic framework of the spatial distribution of the subsidence in your area, which in the end is crucial to properly plan future SAR acquisitions.

In particular, considered that the area you are interested in is affected by a displacement rate which is so large in magnitude (meters per year is really a strong subsidence...), you could exploit also data with longer wavelengths (L-band ALOS PALSAR for instance). This allows to dramatically reduce the problem you would face in vegetated areas due to low interferometric coherence.

 

Q. - Is it possible to monitor Landslides with TerraSAR-X data? Are there special recommendations in terms of acquisition incidence angle?

 

A. - Landslide monitoring with SAR Interferometry is quite a challenging objective, essentially for two reasons: the slope inclination and the possible lack of coherence.

If the slope is not too steep, the first problem can be overcame by properly defining the acquisition look angle; in general we could say that large angles have to be preferred when you observe a slope facing the sensor (to avoid layover conditions), while steeper angles are better when the landslide is located in the shadow side (slope not facing the sensor).

The second problem exists when the landslide area is covered by vegetation; keep into account that even sparse vegetation can cause temporal decorrelation problems (i.e. low coherence) when observing with a short wavelength (X band in your case).

It is important to carefully analyse the slope geometry (orientation, inclination and dimension) before planning the InSAR pair acquisition beam.

If the landslide can be detected and the interferometric coherence is good, you can also plan to acquire an interferometric time series, which eventually allows both to observe the movement with a better accuracy and to better characterize the landslide dynamics by means of the SBAS or PS approach.

A very important issue, especially with VHR SAR data, is the availability of a reliable high resolution Digital Elevation Model. It allows to properly remove the topography during the interferometric processing.