Remote Sensing: The Next Generation
(by Joseph M. Piwowar - published in Cartouche, No. 28, Winter 1998)

At the geomatics conference held in Ottawa last May (GER'97) I ran into Stan Aronoff, author of one of the first - and still widely used - texts on GIS: Geographic Information Systems: A Management Perspective. After the usual greetings and formalities were out of the way, he reached into his briefcase and pulled out his newest book: Total Workplace Performance: Rethinking the Office Environment.

"This seems like a bit of a departure from GIS work," I commented.

"Well ... yes and no," was his reply. "True, office environments aren't generally thought of in a GIS context, but I do have a section in here which describes how you could use a GIS to analyze physical measurements common to an office setup, such as spatially analyzing light levels on a worksurface or mapping the distribution of occupant questionnaire responses."

This exchange prompted me to ask him, "Stan, when are you going to publish the second edition to your classic GIS text?"

He looked at me with a puzzled look, "The fundamental concepts behind GIS haven't changed ... but I suppose the technology and the acceptance of the technology have advanced dramatically."

"Writing a book - even just a newer edition," he continued, "is a lot of work. I put my heart and soul into my books. Besides, everyone is writing GIS texts now. If I was to update my previous book I need to know that I could make a unique contribution. So, Joe ... how can I make a revised edition unique?"

It didn't take me long to suggest an answer, "Remote Sensing: The Next Generation."

"Joe, you've been watching too much T.V.! What does TNG (a.k.a. Star Trek: The Next Generation) have to do with GIS?"

"Not Star Trek," I replied, "but remote sensing! Over the next 5 years there will be several satellite systems launched which will deliver global data at less than 5 metre resolution. These new data sources will revolutionize the way in which spatial data are captured and fed into a GIS. We may no longer need airphotos."

"And you, Stan, are uniquely qualified to write about this," I concluded, remembering his early work in remote sensing.

Whether Stan Aronoff has taken my suggestion to heart, I don't know. But I do know that we are on the brink of a data explosion, the likes of which haven't been seen since the launch of the first routine remote sensing mission, Landsat 1.

A Bit of History

Landsats 1, 2, and 3 (1972 - 1983) carried the multispectral scanner (MSS) instrument which was capable of imaging the earth at 80 metre resolution in 4 spectral bands. At that time, the ability to view the entire earth in 80 m detail was revolutionary. Everybody jumped on the bandwagon. Earth scientists, geographers, environmentalists, urban planners, and even cartographers, all wanted to try out the new data on their favoured applications. By the end of the 70's, though, the honeymoon was over. It was by then pretty clear that although 80 m resolution sounded great, it wasn't going to cut it for many of the applications that people had in mind. Many of the features that these eager pioneers were trying to map dealt with the products of human activity, either directly or indirectly. Many of the details of human endeavour occur in dimensions much smaller than the sensor's 80 m resolution. (I must say that MSS data is still great for some applications, such as vegetation mapping or resource inventories over large, sparsely populated areas.) Interest in remote sensing waned like disco fever.

The next generation remote sensor was the Thematic Mapper (TM) aboard Landsats 4 and 5 (1982 - present), followed closely by SPOT 1, 2, and 3's high resolution vertical (HRV) sensors (1986 - present). At 30 m and 20 m resolution (TM and SPOT, respectively), these data have been proven to be quite useful for a variety of applications, even within inhabited regions. They are still lacking for higher detail work, however, such as urban street network mapping.

Actually, SPOT's HRV sensors work in two modes. First, there is the "multispectral" (colour) mode, in which they collect 20 m data across 3 spectral bands. This was the mode referred to above. SPOT's other operational mode is that of a single band instrument (i.e., "panchromatic" or black & white mode), acquiring data at 10 m resolution. Now we're getting somewhere. Ten metre resolution is good enough to map urban street networks, but still not quite good enough for property mapping.

SPOT's sensors also differ from Landsat's in that they are pointable. Whereas Landsat's sensors only look straight down at the earth, SPOT's pointable sensors allow you to view the same part of the earth from 2 different angles. This can provide stereo imagery which is useful for terrain visualization and topographic mapping. Digital elevation models can be produced from SPOT stereopairs with a mean elevation accuracy of about 5 m: the same accuracy which is inherent in 1:50,000 topographic mapping.

The Next Generation

As 1997 draws to a close, the next generation of remote sensing systems have already begun to assume their places in earth orbit. The Linear Imaging Self-Scanning (LISS-3) sensor aboard the Indian Remote Sensing satellites IRS-1C and IRS-1D (launched in 1995 and 1997, respectively) view the earth at 6 m resolution in their panchromatic modes. At 6 m resolution, the urban road network is well defined and individual buildings are identifiable.

By this time next year, there promises to be no less than 8 new satellites in earth orbit looking down on us at resolutions of 3 m or better; with 6 more to follow by the year 2004 (Table 1). Each of these satellite programs will operate in dual imaging modes, much like the present SPOT system. They will image the earth at their highest resolutions in black and white (panchromatic mode) and acquire colour images (multispectral mode) at a reduced resolution. (The reason behind the differences in spatial resolutions between the 2 modes is simply a limit on how much data a sensor can process. For colour scenes, 3 or more individual images must be acquired simultaneously. If we don't need colour, then only one image needs to be received and more of the satellite's data processing resources can be directed to collecting more pixels to produce a higher resolution image.)

Table 1: High Resolution Satellites Planned for the Next Ten Years

Satellite	Launch Date(s)	Spatial Resolution (m)
Program	(tentative)	Panchromatic	Multispectral
EarlyBird	1997	3	15
SpaceImaging	1997, 1998	1	4
Almaz	1998, 2001, 2004	2.5	4.1
Clark	1988	3	15
IRS-P5, P6	1998, 1999	2.5	10
OrbView	1998	1	8
QuickBird	1988	1	3
SPOT	2001, 2004	5	10
ALOS	2002	2.5	10

What does this mean for GIS?

Over the past decade, GIS specialists have increased their use of remote sensing imagery either directly, as a visual background to other thematic data, or indirectly, as land cover / land use mapping derived from satellite data. I expect this trend to escalate with the new data sources.

Perhaps the biggest markets for this new high resolution imagery will be those dealing with urban applications. High resolution imagery will allow buildings of all sizes, streets, highways, bridges, and railroads to be accurately identified and located. Streets will have a measurable width on the imagery. Centrelines and edge-of-pavement locations can be extracted and updated with existing maps. Building locations and footprints will be measurable. When combined with other vector or tabular data, this information will help urban planners rationalize density and zoning issues. Existing maps will be enhanced by the addition of new visual information about the land itself. It has been suggested that real estate companies could use high resolution images of listed homes to allow prospective buyers a chance to view the property and its surrounding neighbourhood. Are there areas of mature trees and open expanses of lawn around the property? Do the houses have single or double driveways? A satellite image can provide a more up-to-date view of a locality than is possible with traditional aerial photography, and with considerably less image distortion.

Another key point about satellite images which make them unique is that they are collected routinely and repetitively. This will be most useful in another market area: agriculture. High resolution satellite data will intensify the revolution already occurring in the agricultural industry. The buzz-word for today is "precision farming". Instead of blindly spreading blankets of fertilizer or herbicides over entire fields (as they have done in the past), today's farmers want to be able to make their applications only where and when they are needed. Armed with georeferenced, high-resolution field maps processed to show crop vigour, hi-tech farmers will drive their GPS-equipped tractors to the exact locations of the ailing plants and deposit prescribed amounts of chemicals to the soil. Not only does this save the farmer money, but it also saves the environment from excess chemical applications.

Other areas where high resolution satellite imagery will dramatically alter the status quo are:

As with any new technology, expect data costs to be high initially as the first of the new breed of satellites are launched. As more systems are placed into orbit, however, competition will drive the prices down. There will ultimately be some distilling of the weaker companies from the marketplace so don't become dependent on a single supplier for your image data.

The Fine Print

Before I am accused of being overly zealous, let me throw in a few measures of caution. First, you will need to carefully evaluate the capabilities of the imagery against your information needs. The spatial accuracy of the new images may be a vast improvement over what has been available to date, but it may not be fine or accurate enough for everyone. The position of features will have a locational accuracy nominally equal to the resolution of the data you are using. Absolute locational accuracies are probably within 2 or 3 times the data's resolution. This means that if you are mapping street centrelines with 3 m EarlyBird imagery, for example, your average position accuracy will be about 3 m. You can only be guaranteed that your accuracy will be within 10 m. This is not accurate enough for detailed engineering drawings of the sewer network, but is likely sufficient for zoning and census uses. It is certainly as good as - or better than - existing centreline mapping.

Secondly, considerable effort will be required to extract information from the imagery. Using the street centreline example again, there presently does not exist an effective method of automatically extracting this feature. Image users will have to manually trace out the centrelines visually on the displayed scene. This is time consuming and error-prone, but it may be better than any existing method. Another example is in crop monitoring. Areas of crop stress may not be first apparent on the imagery. Image classification techniques requiring user input will have to be employed to delineate these regions. I'm sure the image processing gurus will be able to automate many of these tasks over time, but don't expect to be buying GIS-ready information when you are buying image data. The information will be there, you will just have to work at getting it out.

Lastly, give some thought to how you are going to store and process the imagery. A 25 km by 25 km scene imaged at 5 m resolution will occupy 25 Mb of disk space per band in the image. At 3 m resolution this escalates to 70 Mb. At 1 m, you'd better buy a 600 Mb disk drive.

The Last Word ... or Two

Will the next generation of satellite systems generate the same level of excitement in the mapping community that the launch of Landsat 1 did? You bet!

Is all of this hype at the end of the twentieth century doomed to fizzle into oblivion along with disco and rap? No way!

These satellites are being built and they will be launched. Their high resolution data will revolutionize the way in which many geographic information systems and specialists operate. See for yourself: check out some of the simulated high resolution images which proliferate the web pages listed below.

One final note. You'd better brush the dust off that old remote sensing text sitting at the back of your shelf. You're going to be hearing a lot of people talking about histogram equalizations and maximum likelihood classifications in the months to come.

What to Know More?

Here are some web links for systems referenced in this article:

Also check out the excellent summary on the next generation of satellites compiled by one of my colleagues here at the University of Waterloo, Mike Wulder: http://watleo.uwaterloo.ca/~mwulder/hirespres.html.

Let me know what you think

Now that you know what I think, let me know what YOU think! E-mail me at piwowar@watleo.uwaterloo.ca.

An update to this column was published in Cartouche No. 29.