| GISdevelopment.net ---> GIS for Oil & Gas Proceedings 2000 |
High Resolution Satellite Imagery:From Spies To Pipeline Management
Steve Adam
Director, Satellite Imaging
Canadian Geomatic Solutions Ltd.
1711 - 10th Ave SW
Calgary, Alberta, Canada, T3C 0K1
Steve Adam
Director, Satellite Imaging
Canadian Geomatic Solutions Ltd.
1711 - 10th Ave SW
Calgary, Alberta, Canada, T3C 0K1
Abstract
In the past, high resolution satellite imagery was the domain of national security organizations. However, this has recently changed with the launch of Space Imaging's IKONOS satellite. Launched on September 24, 1999 it is the world's first commercial high resolution satellite, collecting data at 1-meter black/white and 4-meter multi-spectral.
2000 has the scheduled launch of at least two more commercial high resolution satellites. If these satellites are successfully launched, a buyer will be able to acquire imagery every day of the year (barring cloud cover). As an added convenience, an image user no longer has to buy a massive swath of imagery. For example, IKONOS scenes as narrow as 5km (3 miles) can be purchased. This development has opened the door for corridor applications and has been thoroughly and successfully implemented by TransCanada Pipelines in mapping over 1500km of their mainline.
Introduction
Over the past 30 years, Earth observing satellites have collected remarkable imagery of our planet at greater resolution with each passing decade. Today we stand at the pinnacle of civilian satellite technology with the successful acquisition of 1-meter digital imagery from Space Imaging’s IKONOS satellite (www.spaceimaging.com). Launched on September 24, 1999 it is the world's first commercial high resolution satellite, collecting data at 1-meter black/white and 4-meter multi-spectral.
U.S. President Bill Clinton allowed this all to happen by issuing Presidential Decision-23 in 1994 which allows commercial development of high resolution satellites. However, these commercial 'spy' satellites do not come without some regulation. For example, the U.S. government can exercise shutter control during times when they feel national security is an issue and imagery of Israel has to be degraded to 2m.
Presently, the IKONOS satellite is the best source of high resolution satellite image data. However, the Indian Space Agency's IRS-1 C/D satellites (5m resolution) may also be grouped into the 'high resolution' category depending on ones operating scale. With nearly 5 years of archive for North America, IRS-1 image data offers an excellent resource. Less reliable coverage can be found with the Russian 2 m panchromatic SPIN-2 data set (http://www.terraserver.com/). Table 1 outlines a selection present and future high resolution systems.
This year may see the launch of two more high resolution satellites. If these satellites are successfully launched, imagery will be available on every day of the year (barring cloud cover and polar winter darkness). As an added convenience, an image user no longer has to buy a massive swath of imagery. For example, IKONOS scenes as narrow as 5km (3 miles) can be purchased. As will be discussed below, this development has opened the door for corridor applications and has been thoroughly and successfully implemented by TransCanada Pipelines in mapping over 1500km of their mainline.
The new generation of high resolution sensors are also 'high information' sensors. Imagery is acquired at 11-bits per pixel (0-2047 shades), as opposed to the more common 8-bits (0-255 shades), offering a larger dynamic range for image enhancements. For land cover and other environmental applications, multi-spectral information can be extracted from the visible and nearinfrared bands common to these sensors. A further exciting addition is OrbImage's OrbView-4 200-band hyperspectral sensor with 8 m resolution (http://www.orbimage.com/satellite/orbview4 /orbview4.html) and the multi-polar Radarsat-2 SAR with up to 3 m resolution (http://radarsat.mda. ca).
Proliferation of high resolution satellite imagery should see the price of imagery eventually drop from its present, and still reasonable, level (ranging from $12/km2 to over $50/km2 $USD). Furthermore, the recent explosion of internet based distribution and e-commerce can only expand the interest and demand for this new and exciting addition to the pipeline geomatics marketplace.
Table 1: A Selection of Notable Present and Future High Resolution Satellite Missions
- Similar to Landsat-7 bands 1-4 (B 0.45-0.53; G 0.52-0.61; R 0.64-0.72; NIR 0.77-0.88-m)
- Missions last about 45 days
- Off nadir sensor viewing
- Nadir sensor viewing
Study Objective
To establish if high resolution satellite imagery is as effective as ortho-photos identifying population structures (e.g. residences, commercial, industrial buildings) within a buffer of TransCanada's east line right-of-way. Population feature data are captured from digital ortho-photos, IRS-1 satellite (5m resolution), and IKONOS satellite (1m resolution) imagery. Because of their high resolution, clarity, and positional accuracy, the digital ortho-photos provide the baseline data to which all other data sources are compared.
Two issues are specifically explored:
- Classification accuracy: How accurately do feature attributes on the IRS-1 and IKONOS imagery coincide with the feature attributes extracted from the digital ortho-photos ?
- Positional accuracy: How far apart (e.g. distance in meters) are feature attributes on the IRS-1 and IKONOS images from the same features extracted from the digital ortho-photos ?
There are three data sources from which population structures are chosen. Each is summarized in Table 2.
| DATA SOURCE | RESOLUTION | DATE |
| Digital Ortho-photo | 1.0 | Fall 1996 |
| IRS-1 | 5.8 | Fall 1997/98 |
| IKONOS | 1.0 | Jan 20/28, 2000 |
Pilot Study Sites
Three unique segments along the east line are used in this pilot study:
- Low density prairie - total 33km.
- Low/moderate/high density forested - total 10km.
- Moderate/higher density cottages - total 6km.
Features are identified on the data sources through on screen digitizing. Using digital ortho-photos and satellite imagery (ie IRS-1 and IKONOS) in digital format, an operator moves the cursor to select population features. Since the digital imagery is geo-referenced, each population feature selected is inherently assigned a geographic location. A different interpreter is used for each data source to avoid structure identification biases.
Below are the population features/structures identified and compared:
- Residential
- Commercial
- Industrial
- Public use (e.g. hospital, school)
- Outdoor assembly areas (e.g. golf course, campground)
The digital ortho-photos are chosen to provide the baseline because of their high resolution, clarity, and positional accuracy. The interpreter feels that the classification accuracy of the baseline data set is at least 95% correct, for three main reasons:
- The resolution (1.0m) is sufficient to accurately identify the vast majority of population structures. Particularly industrial, commercial, and nearly all residential.
- Updating and editing the baseline data set as comparisons are being done allow 'fine tuning'
- The interpreter's extensive experience in analyzing photography and performing feature identification.
A summary of the results are shown in Tables 3 and 4.
Table 3. Summary Of Positional Results (distance in meters)
| DATA SOURCE | MEAN | ST. DEV | MIN | MAX |
| IRS-1 | 25.8 | 19.4 | 1 | 109 |
| IKONOS | 6.5 | 5.5 | 1 | 44 |
Table 4. Summary Of Classification Results (%)
| DATA SOURCE | CORRECT | MISSED | MIS-CLASSIFIED |
| IRS-1 | 60.5 | 34.8 | 4.7 |
| IKONOS | 91.6 | 5.1 | 3.3 |
Potential Sources of Errors in Structure ID
IRS-1 Satellite Imagery
- Lower resolution than traditional aerial photos. This has a number of effects listed below:
- Low accuracy (about 50%) for high density housing. IRS-1 imagery cannot be effectively used when structures (usually residences) are less than 2 pixels (10m) apart. Trailer parks are particularly uninterpretable due to their high density. For example 39 of the 98 residential structures missed in Dryden were trailers.
- Incorrectly classify structure. Because of low image clarity, many small structures may be routinely classed as residential when in fact they could be commercial or otherwise.
- Inaccurately place identification point. Although correctly identified, low structure definition may cause the interpreter to place an identification point away from the center of a structure.
- Recent changes. Because there is about one year separating the digital ortho-photos (1996) from the IRS-1 imagery (spring/summer 1997 and 1998), some construction has occurred. However, these points were typically classed as 'Extra' and do not impact the 'Total' result
- Image rectification. Positional errors on the order of ±5m can occur simply through the inherent positional errors experienced with geo-referenced image products.
As shown in Table 4, the accuracy of IKONOS imagery is high (over 90%). Although relatively small, there are four elements from which errors may arise:
- Identifying structures disguised by shadow in low sun angle situations. This can be minimized by collecting imagery in the late spring or early fall (if leaf free imagery is desired).
- Interpreting what a structure actually is. This is largely a subjective judgement, unless there is ground verification available. Therefore, this is commonly a source of error in any interpretation that has to be accepted. In reality, the effects of this are small since most structures are easily identifiable residences or larger commercial or industrial features.
- Recent changes. A difference of over three years is clearly evident between the digital ortho-photos and IKONOS data. Over 50 new residences have been added within the town site. As a result the 'Extra' category has little meaning for this comparison.
- Image rectification. Positional errors on the order of ±5m can occur simply through the inherent positional errors experienced with geo-referenced image products.