Conference Agenda

M. Foumelis

The Copernicus Sentinel-1 satellite mission provides global coverage of the Earth’s surface with high-resolution SAR data. Sentinel-1 SLC data and the derived InSAR products have proven to constitute a valuable source of information not only for various mapping applications such as land cover [1], floods [2] and natural hazard damage [3], but also for crop monitoring [4]. However, the processing and analysis of SLC data can be complex and time-consuming, requiring specialized expertise and resources. Several studies addressed this issue with different approaches. Jacob et al. [5] produced Interferometric Coherence data cubes pre-computing all the possible master-slave pairs, resulting in an efficient user experience but with a high overhead in required resources. Ticehurst et al. [6] produced data-cubes of three Analysis Ready Data (ARD) products over Australia: backscatter, coherence and dual-polarimetric decomposition. Kellndorfer et al. [7] produced a publicly available global seasonal Interferometric Coherence data set. Finally, Agram et al. [8] created a workflow to efficiently read and process SLC data accessing single bursts but unfortunately, the implementation is closed source and the results are available only through the Descartes Labs platform.

We propose SAR2Cube as an open framework that aims to make the pre-processing and on-demand computation of InSAR products from Sentinel-1 SLC data more accessible and user-friendly. It uses openEO [9] as the client interface, which supports multiple programming languages, including R, Python, and JavaScript, enabling a wide range of users to interact with, process, and download data.

The desired datacube is a temporal stack of co-registered SLC images. One image, considered as a reference, is used to define the radar coordinate grid where all the others are aligned and resampled. The software used for the pre-processing steps is ESA SNAP. The first required steps are data unzipping and slice assembly, if the Area Of Interest (AOI) is covered by more than one slice. Subsequently the radiometric Calibration process is applied. The final co-registration step is composed by TOPSAR-Split and Apply-Orbit-File on the master and slave images, Back-Geocoding, Enhanced-Spectral-Diversity and de-bursting (TOPSAR-Deburst). Considering the S-1 IW mode, de-swathing (TOPSAR-Merge) is also required only if the AOI covers more than one subswath. Additionally, to produce the differential interferogram products with the on the-fly (OTF) operator, two Interferogram steps are required. Interferogram with geometric components (flat earth and topography) and real and imaginary part for VV and VH interferogram without geometric components that are used to obtain the basis of the geometric components per each one of the images of the dataset. These bases can be linearly combined to obtain all the possible differential interferogram pairs with the OTF interferogram operator. In this step SNAPHU unwrapping module has been used, since the two interferogram must be unwrapped before extracting the geometric component base.
This workaround is the only drawback of the pre-processing step. It is a time-consuming step that can be fixed by saving the geometrical component during the co-registration step.

The resulting stack is composed of all the aligned and calibrated images. For each date, 9 layers are generated: real and imaginary part of VV and VH for backscatter; geometric component base; and, additionally, the longitude and latitude grids, along with the Local Incidence Angle (LIA) and Digital Elevation Model (DEM), are generated only once and will be the same for each date.

In this paper, we present some general aspects of the SAR2CUBE project mainly focused on the differential interferogram and differential phase/coherence generation.

The differential interferogram computation of a dense list, it is the case of Sentinel-1, can be easily and quickly generated thanks to the Python implementation based on XArray [9] and Dask [10] and most of the processes are highly scalable.

Furthermore, SAR2CUBE offer another important feature. Due to the dense time series, it may be impractical to save all the differential phases and coherence of a stack of more than 200 images. In some cases, we can have more than 1000 interferograms. For each interferogram phase and coherence maps must be saved and stored on disk. With SAR2CUBE we can skips this storing process and compute on the fly what we really need. We also can access just a portion of the full processed area through the spatial subset that takes advantage of the geographic transformation matrices and a precise period of data through the temporal subset tool. This information can be then used in a multi temporal interferogram based process, such as Persistent Scatterer Interferometry (PSI).

SAR2Cube is a framework based on re-usable open-source components capable to provide a flexible access to Sentinel-1 SLC data, reducing the barrier for the usage of InSAR products and giving the users the possibility to work with multiple AOIs and parameters interactively thanks to openEO. Additionally, thanks to the Python based implementation of the openEO processes, it is easily extensible with new functionalities.

The European Space Agency is acknowledged for funding SAR2CUBE with the ESA Contract No. 4000129590/19/I-DT - O SCIENCE FOR SOCIE1Y PERMANENTLY OPEN CALL FOR PROPOSALS EOEP-5 BLOCK 4. The European Commission is acknowledged for the financial support within the H2020 MSCA-RISE project HERCULES (grant agreement 778360).

[1] Alejandro Mestre-Quereda, Juan M. Lopez-Sanchez, Fernando Vicente-Guijalba, Alexander W. Jacob, and Marcus E. Engdahl, “Time-Series of Sentinel-1 Interferometric Coherence and Backscatter for Crop-Type Mapping,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 13, pp. 4070–4084, 2020.
[2] Marco Chini, Ramona Pelich, Luca Pulvirenti, Nazzareno Pierdicca, Renaud Hostache, and Patrick Matgen, “Sentinel-1 InSAR Coherence to Detect Floodwater in Urban Areas: Houston and Hurricane Harvey as A Test Case,” Remote Sensing, vol. 11, no. 2, 2019.
[3] Stephanie Olen and Bodo Bookhagen, “Mapping Damage-Affected Areas after Natural Hazard Events Using Sentinel-1 Coherence Time Series,” Remote Sensing, vol. 10, no. 8, 2018.
[4] Dipankar Mandal, Vineet Kumar, Debanshu Ratha, Subhadip Dey, Avik Bhattacharya, Juan M. Lopez-Sanchez, Heather McNairn, and Yalamanchili S. Rao, “Dual polarimetric radar vegetation index for crop growth monitoring using sentinel-1 SAR data,” Remote Sensing of Environment, vol. 247, pp. 111954, 2020.
[5] Jacob, Alexander and Vicente-Guijalba, Fernando and Kristen, Harald and Costa, Armin and Ventura, B. and Monsorno, Roberto and Notarnicola, C., “Organizing access to complex multi-dimensional data: An example from the esa seom sincohmap project,” 11 2017.
[6] Catherine Ticehurst, Zheng-Shu Zhou, Eric Lehmann, Fang Yuan, Medhavy Thankappan, Ake Rosenqvist, Ben Lewis, and Matt Paget, “Building a SAR-Enabled Data Cube Capability in Australia Using SAR Analysis Ready Data,” Data, vol. 4, no. 3, 2019.
[7] Josef Kellndorfer, Oliver Cartus, Marco Lavalle, Christophe Magnard, Pietro Milillo, Shadi Oveisgharan, Batu Osman-oglu, Paul A. Rosen, and Urs Wegm ̈uller, “Global seasonal Sentinel-1 interferometric coherence and backscatter data set,” Scientific Data, vol. 9, no. 1, pp. 73, Mar. 2022.
[8] Piyush S. Agram, Michael S. Warren, Matthew T. Calef, and Scott A. Arko, “An Efficient Global Scale Sentinel-1 Radar Backscatter and Interferometric Processing System,” Remote Sensing, vol. 14, no. 15, 2022
[9] S. Hoyer and J. Hamman, “xarray: N-D labeled arrays and datasets in Python,” Journal of Open Research Software, vol. 5, no. 1, 2017.
[10] Dask Development Team, Dask: Library for dynamic task scheduling, 2016.

Since its first release in July 2018, the open source snap2stamps package has supported a large number of scientists and EO practitioners in exploiting Copernicus Sentinel-1 mission data for measuring terrain motion by means of Persistent Scatterers Interferometry (PSI) [1,2]. The package allows the semi-automatic generation of single master interferogram stacks using ESA SNAP toolbox suitable for further analysis using StaMPS software [3].

Following its public availability on GitHub [https://github.com/mdelgadoblasco/snap2stamps], snap2stamps was downloaded over 5000 times, highlighting the interest of the InSAR community, especially for geohazards applications. As part of official training and capacity building activities, snap2stamps was demonstrated in several international conferences (incl. IEEE IGARSS in 2021 and 2022) as well as in the frame of the Copernicus RUS training service [4].

During these last 5 years, apart from identifying features for successive implementations, new version of several core tools/dependencies were released (e.g. ESA SNAP and python). In addition, interested users contributed by modifying parts of the package according to their needs. Thus, the necessity to evolve the package was underlined.

To address those requirements an evolution of the snap2stamps package is necessary to maintain an undisrupted support to users. In the current work we communicate new features of the upgraded version of snap2stamps (available online since July 2018), among which i) Sentinel-1 TOPS multi-swath processing, ii) support to AOI definition using shapefile, iii) plotting of resampled amplitude images and interferogram phase, iv) resume processing, so the user can stop and resume processing without reprocessing the entire stack at once, v) Jupyter notebooks with usage examples, and vi) a light dockerized Sentinel-1 toolbox. Storage optimization is also part of the upgraded processing scheme.

Apart from the above-mentioned improvements, of importance is the augmentation of the package to support several other EO missions, including TerraSAR-X and COSMO-SkyMed stripmap mode. In this regard, a new package called TSX2stamps was developed by the University of Jena [5], which allows for the semi-automatic generation of single master interferogram stacks using high-resolution TerraSAR-X Stripmap data provided by the German Aerospace Center (DLR) for further analysis in StaMPS. The core functionality is based on snap2stamps, but was slightly adapted for the preprocessing of X-band SAR data, including subsetting, coregistration and interferogram generation using the corresponding SNAP functions. TSX2stamps will also be part of the upgraded snap2stamps version, and the users will be able to use seamlessly the corresponding implementation integrated in the snap2stamps according to the data to be used, snap2stamps for Sentinel-1 data, and TSX2stamps for TerraSAR-X data.

Our goal remains to motivate the users’ community by showcasing the aforementioned major upgrades while inviting domain experts to contribute enhancing and expanding the capabilities of the package.

References

1. Foumelis, M., Delgado Blasco, J.M., Desnos, Y.L. and Engdahl, M., Fernández, D., Veci, L., Lu, Jun and Wong, Cecilia (2018). ESA SNAP - StaMPS Integrated Processing for Sentinel-1 Persistent Scatterer Interferometry, International Geoscience and Remote Sensing Symposium 2018 (IGARSS), 1364-1367.
2. Delgado Blasco, J. M., Foumelis, M., Stewart, C., & Hooper, A. (2019). Measuring urban subsidence in the Rome metropolitan area (Italy) with Sentinel-1 SNAP-StaMPS persistent scatterer interferometry. Remote Sensing, 11(2), 129.
3. Hooper, A.; Bekaert, D.; Spaans, K.; Arıkan, M. Recent advances in SAR interferometry time series analysis for measuring crustal deformation. Tectonophysics 2012, 514–517, 1–13.
4. HAZA09 - SNAP2StaMPS: Data preparation for StaMPS PSI processing with SNAP. https://rus-copernicus.eu/portal/wp-content/uploads/library/education/training/HAZA09_SNAP2StaMPS_MexicoCity_Tutorial.pdf
5. Ziemer, J., TSX2stamps github repository. https://github.com/jziemer1996/TSX2StaMPS

Processing Synthetic Aperture Radar (SAR) imagery is a time-consuming and computation-heavy activity due to large amounts of data and the complex nature of processing algorithms. With new satellites having improved spatial resolution and coverage, and constellations becoming larger over time due to requirements for more timely acquisition of imagery, the data volume keeps increasing significantly over time. To improve the scalability of processing both temporally and geographically, novel methods for SAR processing need to be applied.

A set of SAR processing tools that utilize GPU-s for processing have been developed by CGI Estonia, and consolidated into the ALUs Toolbox software package. The processing algorithms were selected with input from expert organizations in the academia and industry, and are based on equivalent algorithms from the ESA Sentinels Application Platform (SNAP) toolbox. Particular care was taken to ensure that the results of the GPU processing conformed to the results of SNAP processing in terms of quality, and the outcomes were tested in the Amazon Web Services (AWS) environment. The tools implemented so far include the generation of analysis-ready coherence and calibrated intensity products from Sentinel-1 SLC imagery, and focussing of ALOS PALSAR Level-0 imagery. The ALUs software has successfully been deployed and used by the European Commission's Joint Research Center (JRC) in the CREODIAS environment to produce a year-long timeline of analysis-ready Sentinel-1 coherence data to analyze the impact of the Russia-Ukraine war on Ukrainian agricultural activity. Feedback from JRC proves that Sentinel-1 coherence information can be generated in seconds using GPU-s and the outcome of ALUs processing is precise and stable enough to be used for scientific applications.

The latest version of the ALUs Toolbox has been made publicly available and can be found on GitHub: https://github.com/cgi-estonia-space/ALUs. During the latest test, for a full Sentinel-1 swath landmass-only scene, the end-to-end processing time was 15.7 seconds for the coherence estimation routine and 5.8 seconds for the calibration routine. As a comparison, generating a coherence pair using SNAP 8 took around 90 seconds on the same images. Details of the processing routines, and the environments where the processing results were achieved and compared, can be found on the aforementioned GitHub site. It has been identified that the processing speed is heavily affected by the GPU selection, and storage. It has also been identified that significantly better performance can be achieved by GPU-s that support FP64 (double) calculations. Moreover, as storage transfer significantly affects the overall end-to-end performance, a high-performance SSD disk is required to store the data.

The optimization tasks and other improvements are being addressed under an ongoing Estonian GSTP activity. As of early 2023, work is ongoing to support the usage of Copernicus DEM30 and enhance the processing speed even further. There is also an intention to publish the ALUs ARD processors as a public CREODIAS service.

The oral presentation will present the ALUs toolbox's latest achievements, discuss processing speed drivers and accuracy of results when compared to SNAP processors, present the public CREODIAS service and discuss some potential new applications unlocked by the achieved processing acceleration.

Copernicus Programme’s Sentinel-1 SAR constellation images most of the land masses, with a revisit time of 6-24 days, in the Interferometric Wide (IW) swath Terrain Observation by Progressive Scanning (TOPS) mode. The S1 constellation has generated more than 10PB of Level-1 products since September 2014, and the size of this archive is expected to grow 3-4 fold over the next decade as more instruments are added to the constellation. Despite excellent global coverage and temporal sampling, application scientists and remote sensing data users struggle to work with Level 1 SAR data as the data are distributed in non-Geographic Information System (GIS) compatible map projections and the need for custom processing tools to work with these products. With more SAR missions targeting global coverage like NISAR and ROSE-L expected to be launched in the near future, the challenge of making SAR products usable within GIS frameworks to allow a larger community to benefit from these missions will only get more acute.

In this work, we present workflows developed at Descartes Labs that allow users to perform established SAR and InSAR analysis within GIS frameworks. The presented solution not only improves accessibility to SAR and InSAR data, it also allows end users to work with these datasets within the same frameworks as other remote sensing datasets like optical imagery, weather forecasts etc.

Coregistered, geocoded SLC stack

Currently, Level 1 SAR products from various missions are distributed each in their own non-GIS compatible slant range projection systems [1]. Aligning this imagery on a common grid requires specialized processing tools and requires a large amount of computation resources. Distributing coregistered stack of SAR imagery as a Level 2 product will significantly accelerate development of end user analytics workflows and will encourage broader adoption of SAR data in the remote sensing community. We also propose that the coregistered stack is already generated in well known projection systems [1] to allow the large community of users familiar with working on optical datasets to easily adopt standard GIS tools to work with SAR data. We believe a large fraction of end users can easily leverage Level 2 products generated using a DEM chosen for entire missions as is typically done for optical missions like Sentinel-2. Advanced users and experts who require custom processing can always leverage the lower level Level 1 SLC products, as is also the norm in the optical remote sensing community.

Higher level derivative product workflows

Using the Level 2 geocoded SLC stacks as a base product, a number of widely used products can be easily derived within standard GIS frameworks. At Descartes Labs, we have implemented these workflows [1,2,3] and we describe Sentinel-1 specific implementation details.

1. Geocoded SLCs for infrastructure monitoring: For full resolution infrastructure monitoring, we geocode Sentinel-1 bursts to a standardized 10 meter Northing x 2.5 meter Easting grid [1]. The phase of the SLCs are flattened using the same DEM used for geocoding, to simplify further interferometric processing. The real and imaginary values of the complex SLC product are stored as separate bands. This data is accessed in the same manner as bands in optical imagery and time-series InSAR analytics tools have been developed on top of standard GIS frameworks [3].

1. Geocoded Terrain Corrected (GTC) backscatter products: GTC products can be derived from geocoded SLCs using an absolute value band math operation and spatial filtering. Within our data system, we generate GTC products on a standardized 10 meter UTM grid [1] globally from Sentinel-1 IW mode data.

1. On-the-fly Radiometric Terrain Corrected (RTC) backscatter products: We have also developed a formulation to transform GTC products to RTC products on the fly exploiting imaging baseline information similar to InSAR time-series analysis [2]. In the case of Sentinel-1, we have already shown that this transformation can be reduced to a simple band math operation [2] due to its narrow orbital tube. The same framework can also be used to transform GTC products to other calibration levels like (sigma0E or gamma0E) or other types of terrain corrected products [4] on the fly.

1. Pairwise wrapped interferogram products: Pairwise interferograms can be generated from geocoded SLCs by simple cross-multiplication. Interferometric coherence and wrapped phase can be generated from these interferograms using a string of band math and spatial filtering operations on-the-fly. We generate wrapped interferogram products on a standardized 20 meter UTM grid [2] globally from Sentinel-1 IW mode data for all compatible pairs with a temporal baseline of 24 days or less.

We will present some examples of how these derived products can be combined with optical and thermal imagery, on-the-fly to support multi-sensor, multi-modal and multi-temporal analytics.

Mission considerations

We have developed our GIS-based SAR and InSAR processing framework using Sentinel-1 mission as the basis. We believe that the same approach can also be adopted for other medium resolution missions like ALOS-2, NISAR, ROSE-L etc. Finally, we will discuss different factors that one must consider before adopting the proposed framework for large scale processing efforts for these missions, including:

1. Atmospheric propagation delay and its impact on absolute geolocation, particularly for L-band sensors.

2. Accuracy of the Digital Elevation Models (DEM) as we approach ground resolution of less than 2 meters. Adoption of our proposed workflows to higher resolutions over large areas would require global scale DEMs at higher than 10m resolution with a vertical accuracy of less than a couple of meters to be developed first.

References

1. Agram P.S., Warren M.S., Calef M.T., Arko S.A. An Efficient Global Scale Sentinel-1 Radar Backscatter and Interferometric Processing System. Remote Sensing. 2022; 14(15):3524. https://doi.org/10.3390/rs14153524

2. Agram P.S.; Warren M.S.; Arko S.A.; Calef M.T. Radiometric Terrain Flattening of Geocoded Stacks of SAR Imagery. Preprints 2023, 2023020233 (doi: 10.20944/preprints202302.0233.v1).

3. Olsen K.M., Calef M.T., Agram P.S. Contextual uncertainty assessments for InSAR-based deformation retrieval using an ensemble approach, Remote Sensing of Environment. 2023. https://doi.org/10.1016/j.rse.2023.113456

4. Navacchi C., Cao S., Bauer-Marschallinger B., Snoeij P., Small D., Wagner W. Utilising Sentinel-1’s orbital stability for efficient pre-processing of sigma nought backscatter, ISPRS Journal of Photogrammetry and Remote Sensing. 2022. https://doi.org/10.1016/j.isprsjprs.2022.07.023