|
Aerial Photography and Satellite Imagery: Competing or Complementary?
By John Thorpe The author examines the impact which the new 1- meter satellite imagery will have on the aerial mapping industry. Both aerial photography and satellite technologies are briefly described, and satellite orthorectified imagery is compared with conventional aerial digital orthophotography. The advantages and disadvantages of each are discussed, along with issues concerning accuracy and resolution. An attempt is made to table the currently projected costs of both products. Introduction
The launch in late 1997 of the first 1-meter resolution satellite will herald the start of a new era in mapping and GIS. Thereafter, it will be possible to commission the raster imaging of any part of the world, weather and vegetative foliage permitting, within a few days. Ortho-rectified raster maps with limited absolute accuracy will be available within perhaps a week, with high absolute accuracy within a somewhat longer period.
Satellite imagery contains some qualities of resolution and color which are unavailable in aerial photography, and these will make the product more attractive to certain sophisticated users.
Will the aerial mapping industry be severely threatened? The author's answer to that question is an emphatic NO, provided that aerial mapping companies are prepared to meet the challenge. The author believes that the aerial and satellite technologies are highly complementary with one another, and should be made available for sale to the public through the same channels. However, there is certainly a risk that satellite imagery will be oversold, if the public is not aware of the pros and cons of this new technology. In the end, the users will make their own decisions regarding the imagery they will purchase.
In the following discussion, only the raster product of aerial mapping, i.e. digital orthophotographic imagery, is considered. We will call this Aerial Orthophoto Imagery or AOI, and compare this with its satellite counterpart, Satellite Orthorectified Imagery, or SOI. Note that just as AOI is not, by far, the only aerial mapping product, SOI is not the only product from satellites. In this article a mixture of English feet and metric units is used, where the units are commonly used in that particular technology. Imagery Capture
While the final AOI and SOI products may look identical, their respective methods of capture are very different.
Most aerial photography is captured on film (basic resolution about 8 microns) using a conventional 9" by 9" format camera with a 6" focal length, resulting in a 90 degree field of view. The camera is kept approximately horizontal, so the resulting image is equal in scale over the entire area (assuming flat terrain). The flying height is adjustable according to the scale and resolution required for the final AOI. As an example, a common scale for 1"=100' AOI mapping with half-foot pixels is 1"=833" (1:10,000), where the flying height is 5,000 feet above the ground. Stereo imagery is obtained by overlapping consecutive photographs in the same flight line by 60 percent, and by overlapping adjacent flight lines by 30 percent. Large blocks of photography totaling many hundreds of photographs are commonly flown on one day when atmospheric conditions are clear and consistent.
The film negative (or diapositive) is digitized using a high resolution scanner, at a pixel size approximating the final ground pixel. In the example above, this is 15 microns (15 microns times the photo scale of 10,000 is 150 millimeters, or about one half-foot). The scanner records 8 bits (one byte) of radiometric data for each pixel, resulting in 256 levels of gray for a panchromatic image, and a file size of about 230 megabytes per photograph.
Figure 1 shows the geometry of a single aerial photograph, while Figure 2 shows the stereo overlap. SOI capture is achieved using different designs of data capture. In the following discussion, only one such design is described. Other designs use variations based on similar criteria.
Satellite imagery is captured by a camera, digital in nature, which has only a single array of 13,000 sensors, each about 12 microns square in size. The total length of this array is 157 mm (6.2"). The camera's focal length is 10 meters or about 33 feet, so its viewing angle is less than one degree. The camera is aligned so that each line of imagery is perpendicular to the line of flight. It can pivot 40 degrees sideways each way, so it can capture the same angle as an aerial camera, although not all at once. It can also pivot forward to look ahead, or backward to look behind. As it is connected to a very precise GPS and Inertial Navigation System (INS), both the coordinates and the angles of orientation (the exterior orientation) are known for each exposure which consists of simultaneous recording of the 13,000 linear sensors. When pointing straight downward to what is called the nadir point, the scale of the imagery is 1:66,000, the pixel size is 0.82 meters, and the swath width of the coverage of the line of 13,000 sensors is 11 kilometers. At a 40 degree sideways angle the scale is 1:100,000 in the direction perpendicular to the flight line, but still 1:66,000 in the direction of the flight line, resulting in imagery which is affine in scale. The pixel size at this angle is 1.3 meters wide and 0.82 meters long, and the swath width is 17 kilometers. Looking forward or backward from the nadir point further increases these values. All the above numbers are approximate.
With nadir imaging there is almost zero "building lean" in the SOI because of the very narrow viewing angle. This is a distinct advantage over AOI where angles up to 40 degrees are common.
The satellite flies in a sun-synchronous orbit in a plane through the north pole, the south pole and the sun. Each orbit takes about 98 minutes, so it goes around the Earth about 15 times a day. As the Earth is rotating, the orbits "follow the sun," and stay with the sun, so that for the satellite, while on the sunny side of the Earth it is always midday or at least the same time of the day (e.g. 11 a.m.), on the dark side it is always midnight (or e.g. 11 p.m.). Its speed is about 25,000 kilometers per hour or 7,000 meters per second (4 miles per second). That's fast! In order to maintain a steady flow of lines from each exposure of the linear sensors every 0.82 meters, it must fire these sensors about 8,500 times every second. That's very fast! The system collects about 110 million pixels every second. Each pixel in the Space Imaging satellite requires 11 bits per pixel, as against 8 bits for aerial imagery scanned data, so this means that about 150 megabytes of data are collected every second.
The stereo capability of satellites is interesting, and falls into two main scenarios:
Firstly, when the satellite is approaching the project area the camera is pointed forward and the imagery is exposed: subsequently, after passing the nadir point the camera is pointed back, and the area is re-photographed from this different perspective. Hence we have two images of the area, which overlap 100 percent from two different perspectives. However, we must be aware that these two views are very different from the case in aerial photography, due to the fact that the plane of the camera is tilted differently in the two views. It will be similar to the convergent cameras which were constructed many years ago, and dropped because of the difficulty of seeing the stereo picture with conventional methods, because of the distortion of the images. From a stereo operator's point of view, it "tears your eyes apart." Special techniques, unavailable to the users at present, will be needed to extract a DTM, or digitize planimetric detail, from these stereo views.
Secondly, the stereo views can be obtained using photography from two different orbits, possibly on different days. The results will be similar, but the distortion problems will be compounded. It is important to recognize that only one area 11 km wide can be photographed at one time in any particular area. However, the same area can be covered by looking backward, or 90 minutes later during the next orbit, or on subsequent days. About 1,200 km (750 miles) in the flight line direction are photographable during each orbit because the satellite moves so fast, so the stereo imagery must be captured within a period of only two minutes, if done during the same orbit.
The geometry of a satellite camera is shown in Figure 3, and the geometry of stereo satellite imagery is shown in Figure 4. Ground Control
The ground control requirements for AOI are very low compared with a few years ago, due to the availability of airborne GPS (AGPS). As this new technology provides a control point in space at the center of the camera lens at each instant of exposure, accurate to about 10 cm or 4 inches, very little additional control is necessary, perhaps one or two ground control points in each corner of the block of photography. These points are generally paneled before the flight and surveyed at the same time as the flight. Note that maps can be generated without any ground control at all, in theory, but they would be limited in absolute accuracy.
Satellite vendors claim that SOI will be available within hours or days. This may be misleading, as although the imagery will certainly be available (and useful in many cases such as where previous AOI or SOI exists), but to deliver new SOI, more time may well be necessary. SOI control requirements are almost identical with those for AOI, and SOI vendors must have some ground control points before an accurate map can be delivered, and that could take weeks. Just as in AOI, these points must be manually surveyed, identified on the imagery monocularly or in stereo, and checked before the final aerial triangulation adjustment. Aerial Triangulation
Aerial survey companies are beginning to use automatic aerial triangulation techniques. The Helava HATS software and INPHO's MATCH-AT are now becoming available, though to the author's knowledge, no aerial survey company is yet using these techniques on a consistent basis. One must be aware that this technology creates passpoints and tiepoints by image correlation, and "pugs" are not used, so conventional stereoplotters cannot readily utilize the results as there are no "pugged" diapositives. In short, automatic A.T. is useful only in softcopy situations, but as AOI is a softcopy situation, this technique will certainly be used widely in future for AOI.
Satellite imagery must also use "aerial triangulation" to fuse the blocks of imagery together. As the geometry of a strip of imagery is different from aerial photography, the bundle adjustment will use different formulae and algorithms to solve the problem.
Suffice it to say that both AOI and SOI need similar processes to achieve accurate mapping of a large block of imagery. Digital Terrain Modeling
Here again, new technology is becoming available for AOI, notably automatic DTM measurement by image correlation. This technology has been available for years, having been developed largely by Department of Defense contractors especially for satellite imagery. However, existing EOSAT or SPOT imagery generally doesn't require special treatment for forested or urban areas because of the small scale of the imagery. With 1-meter imagery this becomes more important particularly in urban areas, so SOI in this type of terrain will require just as much attention as AOI, because the DTM must be reasonably accurate at ground elevation, and not, for instance on rooftops or at treetop elevation where tall trees are present. SOI will have no problems when nadir or near-nadir imagery is used, due to the very small lens angle, but when using oblique imagery taken at an angle, SOI will have just as much manual editing to do as the AOI providers in order to meet required accuracy. Orthophoto Rectification
For both AOI and SOI, this process is becoming almost entirely automatic. With AOI every photograph is separately rectified, using a space resection to establish the exterior orientation of the photograph (where automatic A.T. is performed this is already available) and then using the DTM to reposition each image pixel in its correct place in the final orthorectified imagery. With SOI, similar rectification procedures must be followed, particularly where non-nadir imagery is used. Orthophoto Mosaicking
This may be an area where AOI will have an initial advantage. Aerial survey companies have invested heavily in proprietary software to merge adjacent orthophotos together into a seamless database where the joins, which were so ugly in the days of hardcopy mosaicked orthophotos, are now almost entirely invisible. Mosaicking is becoming a requirement in GIS where the AOI background to the GIS vectors must not, and need not, show abrupt changes in radiometric values. Satellite vendors have some capability here, and their imagery contains qualities of radiometry which will assist in the mosaicking process, but they may not initially meet the standards of some of the AOI vendors. Accuracy
At least one of the SOI vendors claims that the 1-meter imagery will meet National Map Accuracy Standards (NMAS) for 1"=200' mapping. Presumably this is Class II accuracy, which says that 90 percent of clearly recognizable points must be within 5 feet, horizontally. (Note that Class I accuracy, about two times more stringent, is the proper class for digital products.)
The accuracy of AOI (or SOI) depends for the most part on the accuracy of the DTM. As the angle of viewing increases outward from the nadir point, so any error in the DTM will increase the error in the AOI or SOI.
The USGS DOQQ (Digital Ortho Quarter Quad, where one map covers one quarter of a USGS quad map) program was commenced a few years ago with the objective of covering the entire U.S. with 1"=1,000' scale digital orthophotography with 1-meter pixels. A DOQQ covers 3.75 degrees geographical east and west. The mapping is being produced by six U.S. commercial companies using NAPP (National Aerial Photography Program) 1:40,000 scale photography. Where it exists, the USGS DEM is used; it has a 30 meter regular grid spacing and a declared vertical accuracy of 7 meters RMS, so the 90 percent criterion would be about 11 meters. Figure 5 below shows that at the nadir point, an error of 11 meters produces zero error in the orthophoto, but at a 40 degree angle away from the nadir, an 11 meter error in the DTM would result in a planimetric (AOI or SOI) error of over 9 meters, or 30 feet. This would mean that NMAS standards for only 1"=1,200' scale mapping would be met, not 1"=200'. In practice, the USGS claims that their DOQQ program meets NMAS standards for 1"=1,000' maps, and as very little, if any, of their DOQQ's are mapped from a 40 degree angle, DOQQ accuracy is way within its claimed standards.
Because of the very narrow projection angle of the SOI cameras, SOI obtained from nadir or close to nadir angles and using GPS control points, will, in the author's opinion, fall within the claimed NMAS accuracy. However, imagery taken at angles significantly divergent from the nadir will depend on the accuracy of the DTM used, and will not meet the claimed accuracy unless more accurate DTMs are used.
How will the user know what DTMs were used? Only time will tell. In the author's opinion, the SOI vendors will have to (and probably will be able to) prove their ability to meet these standards under any conditions. SOI vendors will be able to develop their own DTMs to a higher accuracy than the USGS DTMs (provided they have sufficient ground control points). However, that technology has still to be developed and demonstrated to those of us who are somewhat skeptical. Resolution
This is the area where the SOI vendors have superior data to AOI, due to the 11-bit radiometry. (Note: not all satellite vendors will have this extra radiometry.) Satellites also have an additional level of lower-resolution multi-spectral (color) imagery which will be used to enhance the higher resolution panchromatic imagery. However, claims of meeting 1"=200' NMAS accuracy standards should not be confused with meeting the normal resolution standards offered in AOI. What about the resolution standards for 1"=200' mapping? As a matter of fact there are no standards. The American Society of Photogrammetry is currently developing standards for AOI which will include standards for both accuracy and resolution.
The author's company has mapped several AOI projects at 1"=200', where the mapping is done from 1"=1,300' (1:15,600) photography, producing pixels at a size of 1 foot, scanned at a resolution of 20 microns in the scale of the negative. So the users are used to viewing AOI on their GIS projects at 1-foot resolution which has about 10 times the amount of data contained in 1-meter imagery. Once again, only time will tell whether the resolution of 1-meter imagery will satisfy the 1"=200' map users. Clouds, Snow and Vegetation Cover
Aerial photography must be taken in the fall or spring in most parts of the U.S. when leaves are off the trees and there is no snow on the ground. Deviation from these specifications is often cause for rejection by the client. There are parts of the country such as the northeast, where during leaf-on conditions even the roads are invisible from the air because of the large deciduous trees, and any mapping is out of the question for many months. Also, photography must be taken without any cloud cover at all. Not even cloud shadows are tolerated in AOI. Clearly there are many areas of the world, for instance the Amazon basin, when excellent conditions for aerial imaging are almost non-existent. While similar standards must apply to both SOI and AOI, we must remember that SOI vendors will not always be able to produce imagery which is free of snow or vegetation cover. The Market
Customers
The best thing coming out of the new imagery is the tremendous boost which GIS will receive, particularly in desktop GIS, where there are hundreds of thousands of smaller PC-based GIS software packages being used for all manner of business-related GIS projects. These small users don't yet appreciate the value of a layer of landbase imagery, because the proper tools are not yet available. With the huge investment being made by the SOI vendors, these tools will be developed by third parties, and this will increase the demand for both SOI and AOI. The desktop mapping users will become the grass roots of GIS. Sales Channels
The Internet is the sales channel of the future for imagery, at least for small projects. It will be possible to browse through a WWW page, looking at all the available imagery. When a choice has been made, the order will be placed through the Internet, but the data, because of its size, will be sent to the buyer on a CD-ROM, at least for the foreseeable future. Because of the similarity between the final SOI and AOI products, they could and probably should be sold through the same channels. This will occur when strong alliances are created between two industries. Cost
It is interesting to note the difference in costing dynamics between the two producers. On the one hand, AOI costs are mostly variable in nature (i.e. the bigger the area, the higher the cost). Only a few of the costs (mainly aircraft and surveyor mobilization) are fixed costs for any particular project. On the other hand, SOI vendor costs have huge fixed costs (including the satellite and launching costs, as well as ground station maintenance - about $600 million has been publicly acknowledged in Space Imaging's case), while the variable costs of producing imagery are comparatively low.
What does this mean? It may be that the SOI vendors will keep their initial selling rates low to encourage the market to buy their product, because their fixed costs are already in place after the launch. Hopefully for both AOI and SOI vendors, there will be sufficient demand in the future to balance the supply so that an ugly pricewar will not result.
The current cost of rectified imagery varies enormously on a per square mile basis from $1,000 for half-foot pixels to $1 for 30 meter pixels. Clearly, this measure does not provide a fair comparison. Another perhaps fairer method is the comparison of the cost of a screenful of data when viewed with one ground pixel occupying one unit of resolution on the computer screen. A 12" by 15" high resolution screen contains approximately 800 by 1,200 units of resolution totaling about one million units, so a screenful of 8-bit panchromatic imagery contains about one megabyte of data. A strong argument can be made that users will pay for what they can see at optimum resolution on a single screenful.
Calculation of costs per megabyte shows a very different story, with the highest resolution AOI being the least expensive. Table 1 below shows some sample values which have been gleaned from the latest available publications or price lists.
It is interesting to note that one square mile of 30 m EOSAT imagery, costing $1.00, covers an area of only one half square inch on the screen!
So maybe the satellite imagery isn't quite as inexpensive as appears from the current forecasts. And the aerial survey companies aren't standing still. New production methods for AOI are currently being implemented, including automatic positioning of the photographs to facilitate later automation, automatic aerial triangulation, automatic DTM capture using image correlation even in urban areas, and automatic mosaicking of adjacent orthophotos into a seamless database. These new procedures will make the AOI vendors more competitive, perhaps halving the prices quoted in the table above, within the next few years. And if the aerial mapping companies can find a way to sell their product more than once (as the SOI vendors propose to do) they will be able to reduce their prices even further. At this point in time, the only way to directly compare the costs of the two technologies is at the 1-meter level, where AOI is being produced for the USGS DOQQ program. The AOI 1-meter cost quoted above includes the cost of photography and ground control using AGPS, but assumes the existence of a USGS DEM. PROS and CONS SOI will be a wonderful and welcome addition to GIS. The technology is very different from AOI, but the final products (except for the resolution) will be indistinguishable. SOI vendors have more to prove than AOI vendors whose technology is well proven. SOI contains multi-spectral and 11-bit resolution qualities unavailable from AOI. The resolution of SOI may be disappointing at large viewing scales. Clouds and vegetation will be a challenge. Prices of SOI may be less competitive than is being claimed. If a screenful of data is an accurate measure of the true value of rectified imagery, satellite vendors may have to lessen their costs substantially, particularly when they compete with aerial photography companies. Meanwhile, these companies will become more competitive themselves, so the resulting increase in efficiency on the parts of both classes of suppliers will ultimately be of great value to GIS. About the Author:
John Thorpe is chief technical officer and founder of Analytical Surveys Inc. in Colorado Springs, Colo. He may be reached at 719-593-0093 (phone) or 719-528-5093 (fax).
Back
|
