EURAC RESEARCH – ACCADEMIA EUROPEA (IT)
One of the biggest entry level hurdles for integrating SAR data into modern earth observation analysis chains, is still the complexity of the data itself. Opposed to optical or multi-spectral data, which relates much more directly to the way human vision works and is hence more natural to interpret, SAR registers signals that are not natural to human senses, both due to the typically sight looking geometry and to the wavelengths of the electromagnetic spectrum in which SAR typically operates. Operating in this very different spectrum, however, gives the SAR signals unique characteristics, which can be extremely beneficial in many observation scenarios. One of the most obvious and often quoted is the ability to see through clouds and hence SAR is providing a more reliable data source in areas that are frequently cloud covered. Another important domain of SAR are the interferometric observations, which allow for detection of terrain movements or the change of signal strength over time and finally polarimetry, which allows for assumptions on the scatter mechanisms in the physical path of the signal.
SAR2CUBE as a project foresees the development of a processing chain and prototype implementation for organizing Sentinel-1 SLC data in efficient data cubes, both from the point of view of the data provider and from the point of view of the consumer of the data, in order to foster the uptake of SAR data into everyday processing chains. This comprises three main layers, pre-processing, data cube setup and post processing on the fly. A number of different implementation scenarios are planned to be benchmarked in a performance test suite, including necessary space requirements for hosting the data and speed of access from the user perspective. Finally, three use cases have been defined to test the suitability of the developed SAR data cubes for analysis chains, including analysis of terrain motion with PSI techniques, land cover classification based on interferometric coherence and change detection based on back scatter time series
The SAR2CUBE project is defined to satisfy simultaneously two objectives. The first one is to facilitate the use of SAR products in the scientific EO community and to promote them as relevant EO assets. The Sentinel mission within the Copernicus program defines a new playground where to exploit an extraordinary and unique amount of EO information. In particular, the radar pair defined by the twins Sentinel-1A and Sentinel-1B is offering a constant stream of SAR data since they were launched, late 2014 and early 2016 respectively. However, the interferometric capabilities provided by this source are underused. The particular nature of the complex interferometric data often presents a barrier to incorporate these data within the processing chains. The obvious nature of other kinds of sensors, such as optical or multi-spectral data, facilitates the incorporation of these products into different analysis frameworks. To reduce the entry-level barrier of the InSAR-derived products the SAR2CUBE project is designed to provide both SAR and InSAR analysis-ready data (ARD) specifically defined to achieve efficiency and flexibility. Both requirements shall be jointly satisfied to create a reliable and useful framework for the community.
The second objective and equally important is to reduce significantly the amount of storage space that an interferometric SAR context requires. A stack of SAR images responds to a monotonic temporal sequence of acquisitions. Each moment the platform passes by the area of interest a radar acquisition is obtained characterizing uniquely with a timestamp the acquired data. Each image in the temporal sequence is of complex nature, thus they are composed of two bands or channels, the amplitude and the phase. Moreover, if the SAR system has any polarimetric capability the complex acquisition will be obtained simultaneously for different polarisation combinations. SAR interferometry consists of combining pairs of SAR acquisitions. In general, the information contained in the interferogram product is related to the characteristics, evolution and location of the scattering mechanisms in the scene from a relative perspective, i.e. the two images considered. The generation of a useful interferometric stack, multiple pairs of SAR images combined, depends mainly on the application, but the number of possible interferometric products grows quadratically with respect the number of acquisitions, when considering all temporal baselines. To limit the required space for the interferometric context the SAR2CUBE project defines to incorporate a software prototype interface to ARD providing not only access but also on-the-fly processing capabilities.
The performance is very crucial in the whole setup and a balance shall be found between the cost for storage and the cost for computation. Generally, access to analysis ready will be faster if the data is already pre-computed to a higher level, but comes with a higher storage cost, especially in the case of interferometric products.