Got questions? We have answers.
FAQ’s about our products & technology.
How does SkyGeo measure surface movements, from space with millimeter accuracy?
Our products are based on Radar (Synthetic Aperture or SAR) images and interferometric SAR (InSAR). The technology has been around for 2 decades but is now getting applied in infrastructure asset management. Where critical decisions are based on the insights provided by SkyGeo’s products, the decision maker needs to have confidence in the InSAR technique and the way that SkyGeo applies it. Below we list some frequently asked questions about this technology, covering topics such as the satellite radar methods, how the data should be interpreted, InSAR precision and InSAR reliability. We try to answer these FAQs qualitatively and in simple terms. For further questions on these and other topics, please contact us.
How does InSAR differ from conventional measuring methods?
SkyGeo deploys InSAR as a new technique to measure terrain deformations. The technique is opportunistic in nature which is based on radar reflections from infrastructure. Traditional geodetic methods to determine ground movement differ fundamentally from radar interferometry: InSAR measurements are not related to pre-determined physical points on the ground. This contrasts with leveling, tachymetry, or GNSS (GPS) measurements, all of which rely on discrete physical markers such as measuring bolts in infrastructure, underground markers, etc.
Aerial photography, such as shown in Google Earth, is well known and accepted. In fact, these images are formed by reflected sunlight which is collected by a sensor, the camera. Radar images differ from these, because they are acquired in a much longer wavelength signal. The long wavelength enables the signal to travel through clouds. Furthermore, the radar does not rely on reflected sunlight, but acts itself as a source of radiation by transmitting pulses, which reflect on the surface. Finally, the radar measures the time between transmission and reception of the pulse, making it a distance measurement. Aerial photos cannot determine distance and are sensitive to directions under the influence of the lens. This is the essential capability enabling radar images to provide information about distance or distance changes (deformations), while optical instruments cannot.
What are these colored points in the deformation map?
The colored points are radar scatterers. The term comes from scattering, the ability of an object or a composition of objects under the influence of an electromagnetic field, to cause an electromagnetic field itself. In the context of radar interferometry in fact every one resolution cell contains scatterers. The nature of the scattering is highly dependent on the characteristics of the terrain. Thus, a very ‘smooth’ terrain (flat relative to the radar wavelength) causes reflection, leaving little or no return signal towards the satellite. Another area may be “rough” and scatters in all directions. For interferometric applications it is important that the scattering characteristics of an object or resolution cell do not change significantly between consecutive radar images. We call this type of scatterers coherent. If a feature’s reflection remains consistent over time, across many successive radar images, then we call it persistent, hence the term persistent scatterers, abbreviated to PS.
How is it possible that from 800 km altitude, such precision is achieved (millimeters per year)?
This is because the distance is not relevant to the measurement technique. The basis of the measurements is the use of the interferometric phase. Because the phase of the electromagnetic wave can be measured with sufficient precision, it does not matter how long the distance is.
The measurements are ‘relative’. What does a strain rate of 0 mm / year mean exactly?
All measurements must indeed be interpreted relative to one another, as from space is not possible to determine if any point can be considered stable. The interpretation is always “how does point A move relative to point B.”
However, the results need to be shown numerically in the deformation map. For this reason, a reference point is chosen. At this reference point, the strain rate is by definition 0 mm/year. The movement of all other points within the analysis is relative to the (potential) movement of the reference point.
In a product is denoted at point A: -3 mm/year, and at point B: -2 mm/year. The numbers -3 and -2 are dependent on the chosen reference point (i.e. with some uncertainty). If the absolute movement of the reference point is not known, we can only say that point A moves with respect to point B with -1 mm/year.
How is the reference point chosen, and what is the physical interpretation of this point?
The exact location of the arbitrary reference point is not relevant for data processing and the estimated deformation rate. This choice has no effect on the relative uncertainty of the results.
In light of the physical interpretation, however, this choice is relevant. When a moving (e.g. subsiding) point is chosen as the reference point, in our results that point appears to be stable. Without prior knowledge of the physical characteristics of radar reflections, the region, and (geo)physical and geotechnical processes at play, it is impossible to chose an absolute stable reference point. This is a very important factor when interpreting the data, as it easily could lead to misunderstanding!
In order not to get unrealistic deformation numbers we nevertheless try to direct the choice of reference as much as possible. For this reason we choose for example the average of a large number of points in an area considered stable as reference. Consultation with experts from the client may be important in this assumption.
Note that this assumption is identical to leveling, where stable reference points are assumed. Although in case of leveling usually more physical information about the reference point is available, all estimates are based on this assumption.
Can the measurements be made ‘absolute’ to link them to a National Height Reference System (NHRS)?
In theory this is possible through the use of radar reflectors, where the precise location of the radar reflection point is known. However, two variables must be distinguish: (i) the instantaneous absolute height of the point, for example in the NHRS, and (ii) the instantaneous absolute subsidence rate (ie, absolute altitude change of the point relative to the NHRS).
Ad (i) the height of the radar reflection point can only be determined to the level of one meter. The deformation is therefore not estimated by comparing the heights at different times; that accuracy would be too bad. Even when the height of a perfect radar reflector could be connected to the NHRS (by leveling or GPS), the instantaneous absolute heights of all other points would not be available at mm level.
Ad (ii) When the (subsidence) rate from the radar point was also known in the NHRS (ie mm/year, not in mm) then a connection would be possible. In practice, however, (a) virtually no NHRS points whose deformation is determined weekly relative to the NHRS-zero level, (b) the required radar reflectors should guaranteed remain undisturbed for several years, and (c) it is very difficult to make the physical connection between the radar reflector height and the height of the leveling / GPS point.
In short, the value of a link to the NHRS should be critically considered. Just like GPS measurements are made in a reference system (‘datum’) compared to leveling measurements, the radar measurements are referenced – only in a different system. In most cases, however, our measurements are about changes of the points over time.
In what direction does the satellite measure movement?
Using data from a single satellite orbit, we cannot determine the absolute direction of the deformation vector. Only the projection of this vector in the viewing direction of the satellite is measured. However if we do an analysis from different satellite viewing directions, multiple projections are measured. Based on this the absolute direction of the deformation vector can be determined, and we can distinguish (vertical) subsidence from e.g. shear along a slope or horizontal deformation.
What is the effect of the viewing direction of the satellite?
The radar measures only changes in one direction: the line of sight between the satellite and radar resolution cell on the ground. A deformation vector that points in another direction will be projected on the viewing direction, so a portion of the deformation cannot be estimated. By making use of ascending, descending, and adjacent satellite orbits, together with certain assumptions depending on the case, the deformation can usually be estimated in 3D.
The detected points usually originate from hard objects such as rocks, buildings, bridges or other infrastructure. How reliable are these points really?
Very reliable. The technique works so that we can only obtain estimated deformations from reflections which do not change during the observed time interval. However, some major changes of the refleacting surfaces (due to erosion, major maintenance or resurfacing) will result in a change of the radar signal, so no reliable estimate is possible. Precisely the locations where we do not obtain measurements while the conditions (surface) are similar, may be subject to erosion and maintenance. In other words, the radar points are not selected manually by identifying hard objects, but by the criterion of an undisturbed radar signature.
How large is the deformation that we can measure?
This question cannot be answered in general, since the answer depends on several conditions. For example, it is possible to measure changes of several meters with great precision provided that the radar signature of the earth’s surface has not changed. This is i.e. the case for large instantaneous movements (eg, earthquakes or landslides), or large but slow movements in dry, undisturbed areas. In addition, it also depends on the frequency of the radar instrument used. An important element in the discussion is to estimate the phase ambiguities: the number of whole phase cycles in the data. The success rate of correct phase ambiguity resolution depends, among others on (i) the distance between the measuring points in relation to the strength of the deformation signal and the atmospheric signal, (ii) the time intervals between the measurements in relation to the strain rate, (iii) used radar frequency (iv) a priori knowledge of the expected spatio-temporal behavior of the deformation.
What is the horizontal resolution of InSAR?
This depends on the satellite and the radar instrument onboard. In addition, it depends on the look angle of the radar towards the surface. ERS and Envisat have a ground resolution of approximately 5 to 23 meters, and a sampling of approximately 4 to 20 meters. TerraSAR-X has a variable resolution, usually we use 3 by 3 meters. In this discussion it is important to stress that in case of radar images resolution should be interpreted differently compared to optical images. Within a radar resolution cell of 4 x 20 meters, an object with dimensions of less than 1 meter can dominate the reflection. For example, a lamp post covers a much smaller area than 4 by 20 meters, but can sometimes provide a useful reflection.
In the QuickScan you claim to look back in time with this satellite technology. Does that apply everywhere, and how far back can we look?
For this technique, it is important not to work with just one or a few images, but with a large number. This is especially necessary because the distribution of water vapor in the atmosphere, which causes a signal delay, would have a dominant influence on the estimates. Such a series of images must be observed from a similar point in space, so that it is in general not possible to combine different sensors with different orbital parameters. The conclusion is that specific satellite missions should be considered. ERS-1 goes back to 1992, but the satellite was sometimes in a different orbit. This results in ‘holes’ in the time series, which sometimes leads to problems. ERS-2 was launched in 1995 and has been operational until 2011, although the images after mid-2000 were faced with various problems. Envisat was launched in 2003 and recorded a consistent time series until October 2010. After a necessary orbit change, a new time series of images is started at the beginning of 2011 (and will be continued until 2014). However, due to a drift of the orbit, application of these data for deformation studies is limited to certain parts of the world, and one direction. RadarSat-1 has been operational since 1995 and been succeeded by RadarSat-2. TerraSAR-X is operational since 2007. As long as radar images have been recorded, it is possible to analyze the data back in time.
How is InSAR different from the traditional geodetic techniques?
Although InSAR also can be considered as a geodetic technique, there are some differences from the conventional geodetic techniques. For example, the exact location of the measuring point is unknown. We often speak of the effective scattering center, which we do know that the point remains the same during successive measurements.
Another difference is that traditional geodetic networks are designed knowing the purpose of the measurement in advance. Measurement points will be installed where they are needed, given the a priori expectations of the deformation. In case of radar interferometry the network is random, formed by the accidental presence of coherent scatterers. This means that depending on the location, a certain deformation phenomenon is sampled sufficiently or insufficiently. While in case of conventional techniques the physical measurement points can be inspected and can be rejected prior to the measurement, for example because we see that the measurement point is disturbed, or is unrelated to the movement. In case of radar interferometry, this assessment is a posteriori, and has to be derived from the data itself.
Another important difference is that regarding conventional measurements a suspicion of occurring deformation at at particular location must exist. Without such a presumption it would be much too expensive to take measurements. When the radar measurements are performed anyway (as long as image acquisitions are requested), it becomes possible to detect deformation at locations where it was not expected. Other differences are for example the cost of the data, the complexity of the data processing, the decomposition of the deformation vector, and the possibility of independent validation using different acquisition geometries (i.e. satellite look directions).
How are these high-resolution radar images made?
Radar on board of a ship or on airfields is well known. Indeed a radar pulse actually gives one-dimensional information. We send the pulse and then observe the reflections as a function of time. If a point is further away from the radar a reflection will arrive later. The step to a 2D image is simply possible because the satellite moves in its orbit. All civilian radar satellites move roughly in north-south oriented orbits. By combining 1D reflectivity profiles of subsequent pulses, it is possible to create a 2D image. That image looks at first glance like a kind of black and white photograph. However, the principle is entirely different, since distances rather than directions are observed.
What is meant by the (phase) ambiguity in the InSAR measurements?
The interferometric measurements are made by analyzing phase differences of scatterers at different locations and different moments in time. This phase is measured as a real number bounded between the values -pi and pi. Measuring a larger phase difference is not possible because an angle greater than pi will be displayed (wrapped) in the interval between -pi and pi. This creates an ambiguity in the data. These ambiguities in the phase data can be reflected in ambiguities in the estimated height and/or the estimated displacement of the scatterers. Resolving the ambiguity can only be done in a pragmatic-heuristic way, by using differences between points close to each other (assuming that the phase difference between them will be small) or differences between a point observed at different times (assuming the temporal phase difference will be small). If no points can be found relatively close to each other, and/or points with short time intervals are not observed, the reliability of the ambiguity resolution reduces. Note that this does not mean that the quality of the measurement is less. The problem is sometimes solved by using an overlapping radar time series (one location is usually observed from four independent directions) or by including some a priori knowledge of the problem in the estimation.