New perspectives for line map updating
The rigorous SPOT physical geometric modelling allows real-life, accurate photogrammetric applications, like 3D digitizing using stereopairs, and space-triangulation.
Space-triangulation consists in modelling together long SPOT data strips linked to each other through tie points. To compute an accurate and homogeneous physical model through the whole block of images, a few Ground Control Points (GCPs) are necessary, less than ten GCPs per data strip (6 in theory, more in operational situations to reduce GCP errors). These GCPs should be distributed at least at the beginning and end of each data strip. This allows some images having no GCPs at all to be modelled (see Figure 1) which is an undoubted operational advantage. The processing time is also much shorter than a single scene physical modelling which requires the same number of GCPs…but per scene: for instance, a block of 6 data strips of 10 images each, processed using space-triangulation, requires a minimum of around 36 GCPs (6 per data strip). With the same number of images but processed per single scene, it requires around 360 GCPs spread all over the block. This space-triangulation technique allows amazing processing time compared with traditional methods: a block of 250 SPOT images, covering a country of 400,000 km², was processed in around two weeks. The numerous tie points can be acquired, depending on the software, using automatic correlation tools with little human intervention.
The space-triangulation techniques used by Spot Image increases SPOT 5 absolute accuracy: a single SPOT 5 scene has an absolute planimetric accuracy (no GCP) through physical modelling of 50 m RMS. Through space-triangulation, the SPOT 5 absolute accuracy is averaged to 12 m RMS, an accuracy compatible with 1:50000 scale mapping (if relief is correctly corrected) obtained again with no GCP (but with a good DEM). With SPOT 5, the GCPs are used only if they improve the location given by the ephemeris, i.e. if the accuracy of the points is around 12m or better.
So the space-triangulation technique with the SPOT physical modelling potential is really an unrivalled mapping tool for large areas. Space-triangulation has many military applications: it is the only way to obtain a very good accuracy (a few metres) over inaccessible countries, by using long datastrips to reach safe places where it is possible to acquire GCPs (field trips with GPS for instance).
SPOT 5 accuracy
All satellite image providers announce impressive planimetric accuracy for their satellite orthoimages. The problem of this purely commercial approach is that it is not specified that such figures are reached only in some specific conditions, never experienced in operational situations. These specific but not revealed conditions are usually:
- A smooth test area with low relief;
- A place where a huge number of extraordinary GCPs (metric accuracy) and an amazingly good DEM are available;
- Using raw data and satellite ephemeris which are often not provided to normal users;
- Using an exceptional but not marketed software package.
In other words, as soon as the user works with "normal" existing reference data, in mountainous areas for instance, using a commercial software, the geometric nightmare starts.
What one can say about the SPOT system, is that SPOT modelling can reach metric accuracy . The difference from a lot of other sensors is that this photogrammetric potential can be approached much more easily with SPOT in operational situations than with other systems, thanks to what we have seen above: rigorous physical modelling, well known by users and commercial software packages, raw image data and detailed ephemeris provided to customers.
In real situation, the orthoimage accuracy depends on the type of modelling but also on the accuracy of the reference data (a general case for any sensor). The main difference between the SPOT system and other ones is that the SPOT rigorous modelling allows to leverage the best possible benefit from poor existing reference data when other sensors' modelling geometry diverge completely using the same GCPs and show their geometric modelling instability.
In practice, what is the accuracy of SPOT 5 orthoimages? Table 1 gives the accuracy we obtain regularly in relation with the type of reference data used.
Situation number 1 and 2 are reassuring: with no existing ground data but "on-the-shelf" SPOT reference data, it is possible to get reasonable accuracy for our mapping purposes almost everywhere.