Article April 1998.
A New Set of Situations and Challenges.
The advent of the new commercial Earth Observation satellites promising high-resolution imagery from optical sensors is providing a wake-up call to many photogrammetrists who have never previously considered using satellite imagery for topographic mapping applications or the provision of spatial data for input to GIS systems. This article considers both the current situation and some of the implications of this new development from the photogrammetric point of view.
Although space imagery from optical sensors has been available to civilian users on a regular basis ever since the successful launch of the first Landsat in 1972, it has attracted little real interest among those photogrammetrists concerned with topographic mapping. This has been due both to the monoscopic nature of most space imagery and to its relatively low ground resolution. This meant that it was seriously deficient in terms of the features and content that are required even for basic small-scale mapping. Furthermore, until the advent of SPOT in 1986, the stereo-coverage required for the generation of accurate three-dimensional coordinate data was limited to that acquired by photographic cameras during manned space flights of quite short duration. These included the Spacelab missions between 1973 and 1975 and the individual Space Shuttle missions carrying the ESA Metric Camera (MC) and the NASA Large Format Camera (LFC) dating from the early 1980s.The situation changed somewhat after the imagery from the SPOT HRV pushbroom scanner equipped with CCD linear arrays became available. This meant that stereo-coverage of large areas could be acquired cross-track from adjacent orbits on a systematic basis - as distinct from the limited and fragmented coverage that resulted from the short duration photographic missions flown by the Space Shuttle. As a result, a few national mapping agencies began to take an interest in the possibilities of small-scale mapping from SPOT stereo-coverage, mainly in the arid and semi-arid areas around the Red Sea. This resulted in mapping from SPOT stereo-pairs being carried out for part of the Yemen (by Ordnance Survey International) and for Djibouti (by IGN) at the end of the 1980s. Since then, further small-scale mapping has been carried out in this region by the Ethiopian Mapping Authority (EMA) and the Saudi Military Survey Department (MSD) for their respective countries. Most recently, the National Cartographic Institute (INC) in Algeria has embarked on similar project for the small-scale mapping of the Saharan region in the southern part of the country.
Elsewhere in Africa, the revision of large numbers of existing maps using SPOT imagery has also taken place in Nigeria and Uganda. In Canada, part of the national programme of map revision in remote areas has also utilised SPOT imagery. Most recently, the rapid revision and production of new series of topographic maps for the Baltic States has been carried out using SPOT monoscopic images by SSC Satellitbild in Sweden. Other similar projects to revise existing small-scale topographic maps from SPOT images were announced in 1996 to cover large areas of Mexico, Bolivia, Macedonia and Vietnam. Seen from the viewpoint of small-scale topographic cartography, undoubtedly SPOT has been the most successful space imaging programme.
Based on this quite extensive experience, the difficulties that can be experienced using SPOT scanner imagery for topographic mapping are now well known. In particular, its limitations in ground resolution - 15 to 18m in the case of its Pan imagery with a 10m ground pixel size - lead to a substantial part (say 30% on average) of the content of a small-scale topographic map being missing. This leads in turn to the need for a large and expensive field completion programme to remedy this deficiency.
A further difficulty that may occur with SPOT stereo-imagery arises when the individual images making up a stereo-pair are acquired with a substantial time gap between them. The resulting images can then have a substantially different appearance between them due to the changes in land use, vegetation, hydrology, etc. arising from changing seasonal (e.g. summer/winter or dry/wet) climatic conditions. These differences may preclude stereo-viewing and measurement and cause automatic image matching (stereo-correlation) operations to fail.
Similar remarks can be made regarding the imagery from the newer Indian IRS-1C satellite - which was launched at the end of 1995 - and its sister satellite, IRS-1D, launched in September 1997. Each of these utilizes a similar cross-track configuration to that of SPOT for the acquisition of its stereo-imagery, again using linear arrays of CCD detectors, but giving an improved ground pixel size of 5 to 6m.
To overcome the difficulties resulting from the use of a cross-track stereo configuration, the Japanese OPS and German MOMS-02 scanners that were launched in 1992 and 1993 respectively utilize an along-track configuration by which the component images of a stereo-pair are acquired only a minute or so apart instead of months. Thus, for example, the MOMS-02 sensor configuration includes forward and backward pointing linear arrays inclined ±21.6° from the vertical relative to the direction of flight and providing a fixed base:height ratio of 0.8, albeit with a 13.5m ground pixel size.
However, both the MOMS-02 and OPS missions have been conducted on an experimental basis and have resulted in comparatively limited ground coverage. Attempts to provide more extensive coverage have been made by refurbishing and mounting the MOMS-02 system on the PRIRODA module of the Russian MIR space station. This has only had a limited success due to the well documented problems with the MIR station after its collision with a Progress supply vessel in June 1997. Once the repairs have been completed, it is hoped to restart the acquisition of MOMS-2P imagery later in 1998.
With regard to the OPS scanner mounted in the Japanese JERS-1 satellite, this comprises a combination of a nadir pointing sensor and a backward pointing sensor that gives a base:height ratio of 0.3, which rather limits the accuracy of the elevations that can be extracted from the system. In this respect, the forthcoming Japanese ASTER instrument to be mounted on NASA's EOS-AM1 satellite will have a similar configuration, but with the angle of the backward pointing telescope increased to 28° (instead of the 15.3° used with OPS) . The base:height ratio will thus be increased to 0.6, albeit with a 15m pixel size.
In the near future, the whole situation is likely to change drastically with the launch of a number of commercial Earth Observation satellites such as those from Space Imaging EOSAT (IKONOS 1), EarthWatch (EarlyBird and QuickBird) and ORBIMAGE (OrbView) which are designed to acquire high-resolution (1 to 3m ground pixel size) stereo-imagery using both cross-track and along-track configurations. Their main characteristics, together with those of the existing satellites and scanner sensors (SPOT, IRS-1C/D, MOMS, etc.) discussed above, are set out in Table I. Besides the stereo-capability, these systems will deliver imagery of a ground resolution comparable to that of medium-scale aerial photography and, with good base:height ratios, this should allow the accurate determination of elevations for DEM production. This has provided a real wake-up call to the many photogrammetrists who have never previously contemplated using satellite imagery for mapping or map revision purposes.
In practice, space photography is now 15 years old, current interest in this particular field is centered on the short duration (typically 30 day) missions by Russian agencies using space photographic cameras. The main characteristics of the cameras used in these missions are set out in Table II. These missions involve the use of older film camera technology, but this is still capable of producing images with ground resolution values comparable to those of the newer electro-optical scanners favoured by Western countries. Seen from the photogrammetric and topographic mapping points of view, the Russian space photographs, taken mostly with the KFA-1000 and KVR-1000 cameras equipped with f = 1,000mm lenses, are primarily of interest for planimetric mapping and map revision, since their poor base:height ratios and limited stereo-capability precludes the measurement of accurate elevation data.
While the KFA-1000 is a conventional frame camera, the geometry of its photography can be unusual in that it is frequently operated in a twin-camera, split-vertical (i.e. low oblique) mode which is not easy to cope with in most photogrammetric systems. A version of the KFA camera series - the KFA-3000 with f = 3,000mm lens - is also operated in a vertical orientation to give a higher resolution (2m) image. Both of these cameras use a 30 x 30cm format which is too big to be used in analytical plotters or scanners unless the photographs are cut - which is not a very satisfactory solution.
The alternative KVR-1000 is a vertically pointing panoramic camera having the unusual rectangular format of 18 x 72cm with the longer side oriented in the cross-track direction. Thus, as with all panoramic cameras, the image scale will vary considerably in this direction. Confining the use of the image to the central 18 x 18cm part of the photograph reduces this effect and gives a photo scale of 1:220,000 with a 2m ground resolution). The Russian photography has been used to produce satellite image maps at 1:50,000 scale (from KFA photography) and 1:25,000 scale (from KVR photography).
The third main Russian camera is the TK-350, whose configuration is similar to that of the NASA Large Format Camera (LFC), combining, as it does, both reasonably high resolution (7 to 10m) and a good stereo-geometry with a base:height ratio of 0.52. However the difficulties of handling the very large format of 30 x 45cm are a deterrent to many users.
Turning next to consider the photogrammetric instrumentation that can be used with space imagery, the traditional analogue stereo-plotters are little suited to the task since they cannot model or handle SPOT or any other type of stereo-scanner imagery. Even with frame-type space photography, the focal lengths normally used (e.g. f = 1,000mm and 3,000mm with the Russian KFA-1000 and KFA-3000 cameras) are much too large to be accommodated in these instruments. As noted above, a further difficulty concerns the format sizes used with space photographs which are frequently too large to be accommodated in analogue instruments. For example, the Russian KFA-1000 and KFA-3000 cameras have a 30 x 30cm format while a 30 x 45cm photograph is produced by the TK-350 camera. Whereas virtually all analogue stereo-plotters made outside the former Soviet bloc are limited to the standard 23 x 23cm format used by the Wild and Zeiss mapping cameras.
The situation with the alternative type of analytical plotter (AP) using hard copy images is somewhat different. In particular, there are no restrictions regarding focal length, though the large format sizes of the Russian cameras can still give difficulties. Furthermore the AP's control computer can be programmed to model and handle space scanner imagery. Indeed, this has been done successfully for SPOT stereo-pairs in hard copy form with those models in the Kern DSR series using Digital (DEC) computers, plus the the Wild BC2 and BC3 and the Leica SD2000/3000 instruments (using the newer SPOT MS package); the Intergraph IMA (using Trifid software); the Matra Traster (using IGN software) and the NRC Anaplot (using CCRS software). Those mapping organisations that have used SPOT stereo-pairs mentioned above - OSI, IGN, EMA, MSD, INC and the Federal Surveys of Nigeria - have all utilised one or other of these types of analytical plotters with an operator extracting the features and measuring the heights and contours required for their maps from the SPOT stereo-pairs). The Wild Avioplan OR-1 analytical orthophotoprinter has also been programmed to produce ortho-images from SPOT stereo-pairs and indeed is used for this purpose by EMA.
Digital Photogrammetric Workstations (DPWs)
In recent years, the capability of handling SPOT stereo-pairs has been extended to DPWs. This development has come both from the traditional photogrammetric system suppliers and from the suppliers of image processing systems for use with remotely sensed imagery.
Into the former category fall the SPOT software modules developed by photogrammetric system suppliers such as LH Systems (SOCET SET), Intergraph (Trifid) and Autometric/Vision International (SoftPlotter OrthoMAX), all of whom are or have been suppliers of analytical plotters. It is interesting to note that versions of these software packages have also been licensed and developed for sale by remote sensing system suppliers. These include Erdas with the OrthoMAX module and, most recently, PCI with a Windows NT version of SOCET SET to be marketed and sold under the title of Image Works Stereo). The SPOT module of SOCET SET is already in use with SSC Satellitbild, OM&M and Space Imaging EOSAT. The latter two companies also utilize a similar module for the processing of IRS-1C and -1D stereo-imagery that has been developed recently within the SOCET SET software suite. A similar module has just been introduced by PCI. The Trifid SPOT software is used on Intergraph Image Station DPWs by the UK (MCE) and Saudi (MSD) military mapping services
Other remote sensing system suppliers have developed their own digital photogrammetric modules in-house. These include PCI's own EASI/PACE system - which is based on a software package originally developed at the Canada Centre for Remote Sensing (CCRS) - and the MicroImages TNT-mips system. Both of these modules concentrate on automated image matching for the extraction of DEMs and the generation of orthoimages from SPOT stereo-pairs using a high degree of automation. However the lack of stereo-viewing and mensuration capabilities in these modules precludes operator-controlled extraction of features from the SPOT stereo-model. Even more seriously it does not allow the correction and elimination of the inevitable errors and failures experienced during stereo-correlation that need correction by a human operator.
There also exist two or three smaller suppliers whose products are strongly linked with or have been derived from university departments in North America that have teaching and research programmes in photogrammetry and remote sensing. They include R-WEL with its Desktop Mapping System (DMS) originating from the University of Georgia in the USA and DVP Geomatics with its Digital Video Plotter (DVP) developed at Laval University in Canada. Both of these low-cost PC-based systems have modules that can handle SPOT stereo-pairs. So does the VirtuoZo system running on Unix-based SGI work stations which originated at the Wuhan University of Surveying & Mapping in China. This has been transformed into a commercial product and is now being sold by an Australian company based in Brisbane.
The use of stereo-correlation techniques based on automatic image matching for the extraction of DEM elevation data from SPOT stereo-pairs is a feature of all of these DPWs. The matching may be carried out either in the image space or the object space. The corresponding areas to be searched for matches in the two overlapping images may be arranged in image pyramids to speed up the process. Area correlation methods are mainly used, often in conjunction with search strategies that look for matches along epipolar (or quasi-epipolar) lines. Sub-pixel accuracy in correlation can be achieved through the use of least squares matching techniques or by interpolation within the correlation function.
An ortho-image may be generated by geometric rectification of either of the space images making up the stereo-pair or from a single (monoscopic) image - provided of course that a DEM of the requisite quality is available for the area covered by the image. This DEM may have been produced from the stereo-pair itself in a prior operation. Alternatively, the system may use a DEM derived from existing maps that is available and can be purchased from a national mapping organisation - as is the case for many areas in North America and Western Europe. Whatever the source of the elevation data that is used, the final ortho-image can be produced either in digital or hard copy form on the basis of the attitude or tilt corrections generated from the initial image orientation, with the DEM data being used to correct for the relief displacements over the whole of area covered by the image.
The resulting ortho-image can be used to provide a raster backdrop or underlay to an existing vector line map, especially if it is in digital form. The final ortho-image can also be used for monoscopic display and the head-up interpretation and digitizing of point and line features through operator-controlled measurements. This technique has particular relevance and application to map revision.
While users do have the possibility of employing one or other of the packages discussed above with satellite space imagery, enquiries have shown that, for organisations other than national mapping agencies, a good deal of the digital mapping and the generation of DEMs and ortho-images from SPOT stereo-pairs has been carried out by a few third-party contractors having close ties with the image data suppliers. Typical of these are the French companies, ISTAR and CHS, the British companies, NPA and NRSC, and the German GAF company and its Swedish subsidiary, OM&M. Surprisingly these companies are quite often contracted to carry out the work for geoscience users (especially those involved in oil, gas and mineral exploration) and telecomms users, even when the clients themselves have the capability of carrying out the DEM and ortho-image production in-house.
This matter of providing a value-added service or product from space imagery is one that appears likely to come to the fore with the introduction of the new satellite imagery with its 1 to 3m ground pixel size. Some of the data acquisition and supply companies - e.g. Space Imaging EOSAT and its regional affiliates - envisage that much of the product that they will be supplying will be in the form of seamless mosaiced ortho-rectified imagery, DEMs, contour maps, perspective views, fused pan and multi-spectral images, specialized thematic maps and vector and raster data suitable for use in geographic or land information systems. In which case, they will be undertaking both the photogrammetric processing and the thematic interpretation for their clients.
Furthermore EarthWatch has stated that "it will retain ownership and possession of all data collected by its satellites". It also plans to offer a similar range of products from its Digital Globe archive to those offered by Space Imaging EOSAT. ORBIMAGE appears to have a somewhat similar strategy offering a range of standard value-added products through its international network of distributors.
Obviously the first question that will occur to users and potential users is whether the new companies do indeed have the resources and management skills to generate, in a timely fashion, all the value-added products that customers will demand. If indeed, in practice, the supply of value-added products and services does become the major part of these companies' business, then this will be a big departure from present practice where space image data is commonly supplied either in its raw form or with basic geometric corrections only and the customers mainly undertake the further processing themselves in-house employing their own specialist staff. This could become a real issue or, at least, a matter of debate, since not only is space imagery expensive in itself, but a very substantial additional cost will be incurred in the conversion of the image data into map form or some other type of geospatial data - since basically information extraction from any type of imagery is a very expensive operation. This is especially the case if detailed point and line data needs to be extracted for topographic mapping purposes, since this invariably needs extensive human interpretation and measurement.
Thus it remains to be seen how users will react if they find that the costs of purchasing value-added products from the satellite operators and data suppliers is going to be very substantially greater than that of buying raw imagery or imagery with basic geometric corrections. Furthermore one cannot imagine that the photogrammetric and remote sensing system suppliers will be very happy with such a scenario either - since much of their software and expertise will be by-passed.
Virtually all of the new generation of non-photographic Earth Observation satellites are designed to use both along-track and cross-track modes of operation, in combination with linear arrays of CCDs (or areal arrays in the case of Early Bird), very long focal length telescopes and either precisely controlled gimballed mirrors or accurate pointing of the whole spacecraft. Whichever alternative is used, this last feature gives the possibility for flexible pointing (combining both cross-track and along-track pointing) of the imaging sensors towards the required target areas that will often lie to the side of the satellite ground track. This allows both stereo-coverage for elevation measurement and also repetitive coverage for change detection to be generated.
However, at the moment, very little commercial (i.e. non-military) software is available from the photogrammetric and remote sensing system suppliers to allow users to handle along-track or flexibly pointed satellite imagery. LH Systems do now have a module to handle OPS along-track stereo imagery within the SOCET SET software suite running on its DPWs. Nevertheless, at present, much of the photogrammetric software available for the processing and measurement of the along-track imagery acquired by the MOMS-02 and OPS imagers is largely in the hands of universities and space research organisations, e.g. the consortium of German university photogrammetric departments that have developed software to handle MOMS-02 along-track stereo-imagery.
It may be presumed however that the U.S. commercial satellite operators with their prior experience of supplying systems and services to American military reconnaissance and intelligence gathering programmes already have the necessary software and processing capabilities. In this context, one notes that Trifid have developed mathematical modelling for the Space Imaging and EarthWatch sensors which could be used as the basis for the development of suitable software. But obviously the normal photogrammetric and remote sensing system suppliers will all have to develop the capability to handle the new types of satellite imagery if they wish to compete in this market.
What is quite certain is that, with the advent of the new high resolution space imagery, all those sections of the photogrammetric community concerned with topographic mapping will have to cope with a new set of situations and challenges that will have a marked influence on how all photogrammetric operations are conducted in future.
Professor Gordon Petrie can be contacted at the Department of Geography and Topographic Science, University of Glasgow, Glasgow, G12 8QQ, United Kingdom, e-mail:Gpetrie@geog.gla.ac.uk