GLOBAL AUTOMATIC ORTHORECTIFICATION OF ASAR PRODUCTS IN ESRIN G-POD R.Cossu(1), F.Brito(2), O.Colin(1), L.Fusco(1), P.Goncalves (2), M.Lavalle(3) , and M.Paces(3) (1)
ESA-ESRIN Directorate of Earth Observation Program, via Galileo Galilei 2, 00044 Frascati (RM), Italy {roberto.cossu, luigi.fusco}@esa.int (2) Terradue s.r.l, Via G. Peroni, 442/444 00131 Roma,Italy, {fabrice.brito,pedro.goncalves}@terradue.com (3) Università degli Studi di Roma Tor Vergata, Via del Politecnico, 1, 00133 – Roma, Italy
[email protected] (4) IGUASSU Software Systems, Evropska 120, 16000, Prague, Czeck Republic
ABSTRACT This paper presents a simple tool that, thanks to the precise geolocation information contained in Envisat Advanced Synthetic Aperture Radar products alongside with their internal organization, allows a fully automatic and distributed orthorectification of ASAR Wide Swath Medium resolution (WSM) data and ASAR Image Mode Medium resolution (IMM) data. Such a tool has been developed and integrated on Grid Processing on Demand (G-POD). Through the use of a dedicated Web Portal, users can browse for and select ASAR data to be coregistered or to be used for generating mosaics. In both cases, products can be automatically orthorectified. Shuttle Radar Topography Mission (SRTM) Digital Elevation Model (DEM) and GTOPO30 DEM data sets, accessible by G-POD environment, assure the possibility to orthorectify product all over the world. 1. INTRODUCTION Synthetic Aperture Radar (SAR) sensors are becoming more and more important thanks to their ability to acquire measures that are almost completely independent of atmospheric conditions and illuminations. For this reasons SAR data can play an important role in several applications including risk assessment and management (e.g., landslides, floods), environmental monitoring (e.g., monitoring of coast erosion, wetland surveying, forest surveying), etc. [1] From a methodological point of view, different factors render the analysis of SAR data a challenging task, such as the presence of distortions due to the nature of a SAR sensor and to its imaging geometry (e.g., presence of shadow, layover and foreshortening phenomena). When a Digital Elevation Model (DEM) of the observed geographical area is available, a procedure called orthorectification can be applied allowing the correction of some geometric distortions. SAR orthorectification [2]-[5] usually requires four steps: i) a SAR image is simulated based on available DEM and satellite orbital parameters; ii) simulated image and original products are coregistered (in a manual or automatic way); iii) each pixel of the (coregistered) original image is associated to topographic height information; iv) height information is used to correct geometric distortions. Recently, ESA EO Science Application Department i) _____________________________________________________ Proc. ‘Envisat Symposium 2007’, Montreux, Switzerland 23–27 April 2007 (ESA SP-636, July 2007)
has demonstrated how a Grid infrastructure can respond to the complexity and constraints imposed by applications in EO domain and ii) has identified the benefits of this technology and how it can improve the work of EO technical and scientific users. A Grid environment, in fact, can facilitate interactions between different actors by providing a standard infrastructure and a collaborative framework to share data, algorithms, storage resources, and processing capabilities. Furthermore, ESA proposed to facilitate Grid resources and services access through user-friendly application portals. ESA developed a grid-based system, called Grid Processing on Demand (G-POD)[6],[7], in which various EO applications (and related Web Portals) have been integrated. Among these, services related to the analysis of SAR data are now available for internal use. This paper presents a new tool that allows a fully automatic and distributed orthorectification of Envisat ASAR WSM and IMM data. Automatic capabilities have been obtained thanks to the precise geolocation information contained in ASAR products; distribuited features have been developed thanks to the structure of ASAR products that can be easily divided in several processing blocks [8]. Such a tool has been integrated on G-POD and is able to fully exploit the underlying grid infrastructure. The paper is organized as follows. Section 2 gives an overview of the G-POD environment. Section 3 describes the proposed orthorectification approach, and Section 4 details its integration on G-POD. Examples of orthorectification results obtained in G-POD are reported in Section 5. Conclusions are drawn in Section 6. 2. GRID PROCESSING-ON-DEMAND Even though the description of G-POD is out of the scope of this paper, in this section we provide a brief overview of this environment. The ESA Science and Application Department of Earth Observation Programmes Directorate at ESRIN has focused the attention on the development of a dedicated Earth Science Grid infrastructure, under the name Earth Observation Grid Processing on-Demand [6],[7]. This generic Grid based environment (G-POD) ensures that specific Earth Observation (EO) data handling and processing applications can be seamlessly plugged into the system. Coupled with high-performance and
sizeable computing resources managed by Grid technologies, G-POD provides the necessary flexibility for building a virtual environment that gives applications quick access to data, computing resources, and results. This makes G-POD an ideal environment for processing large amounts of data, developing services which require fast production and delivery of results, comparing approaches and fully validating algorithms. Using a dedicated Web Interface, each application has access to a catalogue like the ESA Multi-mission User Interface System (MUIS) and storage elements. It furthermore communicates with the underlying Grid middleware, which coordinates all the necessary steps to retrieve, process, and display the requested products selected from the large database of ESA and third-party missions (see Fig.1). The integration of Web mapping and EO data services using a new generation of distributed Web applications and the OpenGIS specification [9] provided a powerful new capability to request and display Earth Observation data products in a given geo-temporal coverage area. At present, the ESRIN controlled infrastructure has a computing element (CE) of more than 150 PCs, mainly part of four clusters with storage elements of about 100 Terabytes, all part of the same Grid LAN in ESRIN, partially interfaced to other Grid elements in other ESA facilities such as the European Space Research and Technology Centre (ESTEC), the European Space Astronomy Centre (ESAC), and EGEE. The key feature of this Grid environment is the layered approach based on the GRID-ENGINE which interconnects the application layer with different Grid middlewares. In computational terms, the GRIDENGINE is an application server accessed by SOAP Web services. On top of this, the GRID-ENGINE allows the definition of simple service chaining (more in the line of information flow) where the services can be stitched together with their results being automatically defined as input parameters for the subsequent services. Each G-POD EO application accesses the available Grid resources and services through a dedicated Web Portal. With the necessary variables requested by the user and the parameters defined by the application manager for the actual service, the Web Portal will send to the GRID-ENGINE all the necessary information for the instantiation of all templates defining the service. Among the others, a portal dedicated to the analysis of SAR products is available (see Section 4 for more details). Through this portal, users can browse for the required products specifying the geographical area of interest as well as the acquisition time, and, if required, limiting the search for a given mode or pass; perform accurate coregistration or generate mosaics with the selected products [10].
3. THE PROPOSED ORTHORECTIFICATION APPROACH Given the nature of the sensor, SAR images are affected by serious geometric distortions [11]. Fig.2 represents the geometry of topographical distortions in SAR imagery. Being the ASAR squint null, topography leads to a displacement of pixels along the range direction. As an example, a point B at elevation h (topographic height) above the reference ellipsoid is imaged at position B ′ (please note that the range distances R of the two points is the same), though its real position is B ′′ . The offset Δr between B ′ and B ′′ exhibits the effect of topographic distortions.
Figure 1 . The architecture model for G-POD. When a Digital Elevation Model (DEM) of the observed geographical area is available, a procedure called orthorectification can be applied allowing the (partial) correction of the above distortions [2]-[5]. SAR orthorectification usually requires four steps: i) a SAR image is simulated based on available DEM and satellite orbital parameters; ii) simulated image and original product are coregistered (in a manual or automatic way); iii) each pixel of the (coregistered) original image is associated to height information; iv)
height information associated to each pixel is used to locally correct the distortion (i.e., the offset Δr is compensated). Thanks to precise geolocation information contained in the header of ASAR IMM and WSM products [4], we herein proposed a simple fully automatic algorithm composed of few steps, which doesn’t require SAR image simulation or simulated image to original product coregistration. Let X and X ortho be a ground range image corresponding to a medium resolution ASAR product and its orthorectified version, respectively. Let x(i, j ) ( xortho (i, j ) ) represent a generic pixel at row i, column j in X ( X ortho ). Let x (i, j ) ( x ortho (i, j ) ) be the value (e.g., intensity) associated with the pixel x(i, j ) ( xortho (i, j ) ). Let lat i , j , long i , j , R(i, j ) , and η (i, j ) be the latitude, the longitude, the slant-range distance and the local incidence angle corresponding to the pixel x(i, j ) , respectively. Let hi , j = h(lat i , j , long i , j ) represent the elevation
tangent to the ellipsoid at position lat i , j , long i , j and is given by: Qi , j = R(i, j ) cosη (i, j )
(3)
with s being the pixel size in range direction. Eq. 2 is obtained with simple geometric considerations under the reasonable assumption that the incidence angle η (i, j ) and the angle between the line connecting the sensor to the observed point and the perpendicular to the ellipsoid at position lat i , j , long i , j are similar (see fig.2). As usually Δri , j does not assume an integer value, liner or spline interpolation is used in eq. 1. The developed tool also produces shadow maps and layover maps [13]. Furthermore, the tool allows the user to perform (if required) empiric radiometric correction based on the local incidence angle [14], and to correct the antenna pattern from the elevation effects, particularly evident in WSM products [15]. Fig.3 shows a detail of an ASAR IMM product acquired on January, 11th 2006 over Mt Etna, Italy, before and after the proposed automatic orthorectification.
corresponding to the geographical coordinates lat i , j , long i , j .
For each pixel position coordinate (i, j ) in X ortho , the orthorectification algorithm is composed of the following steps: 1. The values of lat i , j , long i , j R(i, j ) , and η (i, j )
2.
are computed. Product header contains accurate information related to the above values for a reduced (but significant) number of pixels in the image, called tie points. Values corresponding to other pixels are estimated via Delaunay triangulation [12]; given the geographical coordinates lat i , j and long i , j , elevation
3.
information hi , j
from the available DEM with interpolation; compute x ortho (i, j ) as follow: xortho (i, j ) = x (i, j − Δri , j )
is derived a
bi-linear
Figure 2. Geometry of topographical distortions in SAR imagery. A point B at elevation h above the ellipsoid is imaged at position B ′ , though its real position is B ′′ . The offset Δr between B ′ and B ′′ exhibits the effect of topographic distortions. Eq. 2 is derived under the assumption that the two angles η and θ are similar.
(1)
where Δri , j is a slant range displacement estimated taking into account the geometry of topographical distortions in SAR imagery (see Fig. 2): 1 Δri , j ≅ Qi , j − hi , j tan η (i, j ) − s
[(
)
2 ⎤ ⎛ Qi , j − hi , j ⎞ ⎜ ⎟ − Qi2, j ⎥ (2) ⎜ cosη (i, j ) ⎟ ⎥ ⎝ ⎠ ⎥⎦ where Qi , j is the sensor’s distance from the plane
a)
b)
Figure 3. detail of an ASAR IMM product acquired on January, 11th 2006 over Mt Etna, Italy, ascending pass. a) before orthorectification, b) after the proposed automatic orthorectification.
4. INTEGRATION OF THE ORTHORECTIFICATION ALGORITHM IN GPOD The developed orthorectification tool has been integrated in G-POD. Through the SAR G-POD portal, users can decide to orthorectify products, for example when generating ASAR mosaic. Thanks to this tool and different SAR toolboxes previously integrated on G-POD, it is possible to perform fully automatic SAR image despeckling, backscattering computation, image co-registration, flat ellipsoid projection and topographic correction for medium resolution images, and mosaicking. Higher level functionalities have been developed and made available through the SAR G-POD web portal. Users can browse for the required products specifying the geographical area of interest as well as the acquisition time, and, if required, limiting the search for a given mode or pass. Afterwards they can specify the service of interest (e.g. co-registration of multitemporal images or mosaics) (see Fig. 4). The system automatically retrieves data stored on different storage elements (e.g. distributed archive), identifies the jobs needed for accomplishing the task required by the user, and distributes them on different computing nodes of the Grid. Concerning DEM dataset, the Global Shuttle Radar Topographic Mission (SRTM) DEM v3 (which has a resolution of 3 arcsec, comparable with the one of ASAR medium resolution products) was downloaded from the U.S. Geological Service web site [16]. For latitudes above 60 degrees north and under latitudes 56 south, GTOPO30 DEM [17] was downloaded. Both DEMs were stored in different storage elements of ESRIN Grid infrastructure (the total size of the DEMs is approximately 120Gbytes). Necessary DEM tiles are automatically identified and retrieved (in a fully transparent way) from the storage elements. As discussed in the previous section, in the proposed procedure, each pixel of the product is automatically associated to height information in the DEM and the geometric distortion is corrected. As SRTM DEM contains some holes (area for which height information is not available) and EGM-96 geoid is used as vertical datum [18], a preprocessing phase is performed in which holes are filled using Delaunay triangulation and vertical datum is converted to WGS84. It is worth noting that internal organization of ASAR products (which are structured in units called granules) allows a distributed orthorectification of Envisat ASAR data. The tool is developed in such a way that user can decide to orthorectify different granules of the same product on different working nodes of the Grid (Fig.5).
Figure 4. The Web Portal for the ASAR G-POD environment.
Figure 5. Thanks to the internal organization of ASAR products, the orthorectification of a single product can be distributed among different computing elements of the grid infrastructure. 4. RESULTS The integration of this orthorectification procedure in the processing chain of higher level ASAR end-to-end services in GPOD, allows new interesting scenario. In particular, we identified three services that can notably benefit from the developed procedure, i.e., mosaicking, analysis of multitemporal data, and data fusion. As regard to mosaicking, the new functionality makes it possible to produce global or continental scale high resolution mosaics of orthorectified radar products. As an example, the G-POD SAR processing application produced a 3 arcsecond (~90 meters) pixel size orthorectified Envisat ASAR mosaic over Europe (see Fig. 8). The mosaic was generated using ASAR WSM data acquired between January and May 2006, the Digital Elevation Model (DEM) at 3 arcseconds derived from the global Shuttle Radar Topography Mission (SRTM) and the GTOPO30 DEM (for latitudes above 60 degrees north). The whole process was achieved in few hours. To cover the whole of Europe, 143 ASAR WSM stripline products were automatically selected and
retrieved from G-POD storage together with required SRTM and GTOPO30 DEM tiles. Products were orthorectified and normalized for near-range/far-range effects over sea before being aggregated in a mosaic. The so obtained result was used to produce the SAR European mosaic poster distributed at the Envisat Symposium 2007 [19] and used as ESA Image of the week on March 16th ESA Web Portal [20]. Concerning the analysis of multitemporal data, when monitoring a given geographical region, orthorectification allows considering products acquired with different passes (i.e., ascending and descending) and different incidence angle, so virtually increasing the temporal coverage of the regions. Important applications include flood monitoring (see Fig. 6). Finally, orthorectification is the first step when merging SAR data and optical data to address complex classification problems. In Fig.7, we show a multisensor composition obtained merging Landsat TM data (derived from the landsat WMS mosaic [21]) and an orthorectified ASAR mosaic. All the processing (except for Landsat mosaicking) has been automatically performed in G-POD. 5. CONCLUSIONS A simple, yet effective, orthorectification tool has been developed for ASAR medium resolution products. Thanks to the precise geolocation information contained in ASAR product this tool is fully automatic. Despite the approximations assumed in the algorithm, experimental analysis (not reported in this paper) demonstrated a good accuracy of the results. The integration of the tool in G-POD, alongside with the availability in this environment of SRTM DEM and GTOPO30 DEM dataset, allows the global orthorectification of ASAR products. Through dedicated web portal user can easily browse for products to be mosaicked or coregistered and may decide to orthorectify such data. Required DEM are automatically retrieved and used for the processing. As a result, this tool together with other toolboxes already integrated in G-POD allows the generation of global scale high resolution orthorectified mosaics, accurate coregistration of products, multisensor data merging. At the present, new tools able to orthorectify IMP products are under study. Acknowledgment We are very grateful to all members of the ESA EO grid team, especially to E.Mathot, and V.Forneris, who have directly contributed and impacted to the design, implementation, validation, and operation of G-POD services. Finally, we thank C.Retscher for the fruitful discussions on processor integration matters.
Figure 7. Multitemporal composition obtained by orthorectifying and coregistering a time series of WSM images acquired over Chinese Poyang lake.
Figure 7. Multisensor composition obtained by orthorectifying and merging ASAR WSM images and Landsat TM data acquired over Geneva lake, Switzerland. 6. 1. 2.
3.
REFERENCES http://envisat.esa.int/object/index.cfm?fobjectid=37 72&contentid=3804 J. C. Curlander, R. N. McDonough, “Synthetic Aperture Radar: Systems and Signal Processing”, Wiley Series in Remote Sensing, 1991, ISBN: 0471-85770X. Y.Sheng and D.E.Alsdorf, ”Automated georeferencing and orthorectification of Amazon Basin-wide SAR Mosaics using SRTM DEM
Data”, IEEE Trans. Geoscience and Remote Sensing, vol43, 8, 2005, pp1929-1940 4. G.M. Huang, J.K. Guo, J.G. Lv, Z. Xiao, Z. Zhao, C.P. Qiu, “Algorithms And Experiment On Sar Image Orthorectification Based On Polynomial Rectification And Height Displacement Correction”, Proc. of Isprs, 12 -23 July 2004, Istanbul, Turkey. 5. Leland Pierce, Josef Kellndorfer, Fawwaz Ulaby, 1996. “Practical SAR Orthorectification”, IGARSS, 4, pp. 2329 –2331. 6. http://eogrid.esrin.esa.int/ 7. L.Fusco, R.Cossu, C.Retscher, “Open Grid services for Envisat and Earth observation applications” in High Performance Computing in Remote Sensing, Ed: Antonio Plaza, Taylor and Francis Group, Chapter 13, in press 8. http://envisat.esa.int/pub/ESA_DOC/Envisat/ASAR /asar.ProductHandbook.2_2.pdf 9. http://www.opengis.org. 10. L. Fusco, P. Goncalves, F. Brito, R. Cossu, C. Retscher (2006): A new Grid-based system to assist users in ASAR handling and analysis, European Geosciences Union General Assembly Vienna, Austria, 02 – 07 April 2006. 11. F. M. Henderson, A. J. Lewis, “Manual of Remote Sensing, Principles and Applications of Imaging Radar”, Wiley, 1998, ISBN: 0-471-294063.
12. Mark de. Berg, “Computational Geometry: Algorithms and Applications”, Springer, 2000, ISBN 3-540-656200. 13. W. G. Kropatsch and D. Strobl, “The Generation of SAR Layover and Shadow Maps From Digital Elevation Models”, IEEE Transactions On Geoscience and Remote Sensing, Vol. 28, No. I . January 1990. 14. D. Small, E. Meier, D. Nuesch, “Robust radiometric terrain correction for SAR image comparisons”, IEEE International Geoscience and Remote Sensing Symposium, 2004. IGARSS '04. Vol. 3, 2004 Page(s):1730 – 1733. 15. F. Holecz, A. Freeman, J. van Zyl, “Topographic effects on the antenna gain pattern correction” Geoscience and Remote Sensing Symposium, 1995. IGARSS '95. Vol. 1, July 1995. 16. http://srtm.usgs.gov/ 17. http://edc.usgs.gov/products/elevation/gtopo30/gtop o30.html 18. http://cddis.nasa.gov/926/egm96/egm96.html 19. http://www.envisat07.org/ 20. http://www.esa.int/esaEO/SEM9QLQ08ZE_index_ 0.html 21. http://onearth.jpl.nasa.gov/
Figure 8. Three arcsecond (~90 meters) pixel size orthorectified Envisat ASAR mosaic obtained using G-POD. Political boundaries have been manually overlaid. The full resolution result can be seen at [16].
