The new view of Earth.
This section is an extract from "Looking back" by Graetz et al 1992 . It is an overview of some of the principles of interpeting satellite imagery.
What is special about satellite images?
My answer is everything.
Satellite images are really something different: They are unique and immensely powerful in conveying information about the environment.
Satellite images are powerful not in themselves but in the way that they empower every one of us who has the opportunity to use them. To examine a satellite image of your own city, town or farm, is to experience something that virtually no one else has done. You become one of the very few; a genuine minority member. You have experienced a new and potent view of your country.
Genuine hands-on access to satellite images was very limited until quite recently. This restricted access is the consequence of a lack of awareness and easy public availability. I believe the lack of access to satellite images by the Australian public has crippled contemporary environmental debate.
What is the best way to describe the view of life on Earth that satellite images can provide?
It is difficult to find the words to convey my enthusiasm and excitement for satellite images because so many of the powerful words that were once reserved for special occasions are now used for the ordinary and the commonplace.
Let me start by admitting that each time I examine a satellite image, I am excited by and engrossed in what I see. No matter what part of Earth is captured by the image, I am enthralled by the information that space images convey to me, and captivated by their novelty and magnificence.
My response to space images is not only from my head. It is not just the information content of space images that excites me. My sense of delight is also involved because the view of Earth from space is intrinsically beautiful and enchanting.
I have had this affair with satellite images for two decades. It began with love at first sight in 1972. My passion remains as strong as ever. I still can recall the excitement of the first encounter, and the feeling of disbelief at what was captured by a lifeless machine onboard a spacecraft 900 km out in space that was (and still is) ceaselessly circling the planet.
I know that I am not alone in my feelings. I know of many others who also appreciate the beauty of space images or who use their power to communicate ideas and information. I am very confident that I can convey at least some of my enthusiasm for the space view of Earth and convince you of its potential.
Let me begin to infect you with this excitement and sense of wonder by examining the following satellite image.
This is a marvellous image of part of the Simpson Desert, NT. This space image is approximately 30 000 km2 in area and it captures how this remote part of Australia looked from 900 km out in space on June 7, 1981. There is spectacular and intriguing pattern within this image. The patterns of colour and texture are the consequence of natural processes that have been operating in these landscapes over the last million years or so. To me, the most eye catching patterns are those which are the most recent; they were generated in only the last ten years or so. Are you able to interpret the image and identify these patterns as well as speculate on their causes? This image was deliberately chosen to be strange and mysterious, although the details of the landscape are still discernible.
Details: The image identification is WRS 107-077, 22328-00052, 07/06/81. The image size is 91.5 km by 86.2 km, or ~7, 890 km2.
The next image does not offer quite such a synoptic view , but it is spectacular nonetheless, and instantly recognisable.
It is an image of a Dry Season wildfire in Arnhem Land, NT, as recorded by an imaging system called the Multi-Spectral Scanner (MSS) onboard spacecraft number 5 of the Landsat satellite series at 0930 hours AEST on September 2, 1986. This looks just like a fire viewed from an aircraft. Remember, this image was recorded 900 km above the surface of Australia.
Details: The image identification is WRS 104-069, ID 58267-00414, 02/09/86.
If you examine the image closely, you can discern a great deal of detail about the fire: the actively burning fire front, the many smoke plumes, and the colours and shapes of the unburned landscape as well as the fire footprint (scar) left behind as the fire advances.
I still find it an amazing realisation that this scene was recorded by an imaging device onboard the Landsat spacecraft, which is the size of a Volkswagen sedan car, orbiting the Earth some 900 kilometres above that fire.
The steps in the process were these. First, sunlight was reflected from the fire through the atmosphere to the spacecraft where it was optically captured and converted to a coded electrical signal. This signal was in turn converted to a radio signal that was transmitted back to Earth, received at Alice Springs and converted back to an electrical signal. The very long sequence of electrical signals that comprise this image was stored as numbers on a magnetic tape. I purchased a copy of that magnetic tape and, using a computer and other hardware, converted the electronically coded numbers into light signals that were focussed onto optical film. The result is the image you have just examined.
By using relatively simple technology we can in effect put ourselves into the Landsat spacecraft to obtain this remarkable view of the Earth.
This view is an awe inspiring panorama, the like of which is entirely unobtainable on Earth.
What is more, the image that we are analysing will never die.
What was happening on the ground in Kakadu National Park on the morning of September 2, 1986 is stored on magnetic tape in several locations. Forever into the future, this image can be recalled and reanalysed whenever and wherever anyone is interested.
With these thoughts in mind, what else can be interpreted from the wildfire image? What other information can be extracted?
With a little thought and no specialised knowledge, it is possible to infer other details of this wildfire and of the environmental conditions at 0930 hours AEST on September 2, 1986. For example, it is possible to determine which direction is North and then decide the direction as well as the speed of the prevailing wind on that morning. The second answer follows from the first. The first answer can be arrived at by recalling the time of Landsat overpass, then deducing where in the sky the sun will be. The location of the shadows of the smoke plumes is useful here.
After answering these questions, at least two others become obvious: What size is this fire or what is the scale of the image and why is the smoke grey-white while the land surface is yellow-red?
To satisfactorily answer these and other questions about the nature of satellite images, it is useful to step back from the detail and think about this topic in a general way. What is special about the view of Earth from space?
This question began this section; the title of which suggests that the view from space is a new view. That is correct. It is a new view but it is new in two ways: it is new in time and it is new in experience. The novelty to the human experience is the more important.
The view from space is above all; an un-human view. It represents a perspective of planet Earth that was until 1972 absolutely external to the human experience throughout our entire existence as a species.
Besides the attribute of newness, in what other ways does the view of Earth from space differ from the everyday one of human experience? Four differences are particularly significant and critical to developing an appreciation of, and familiarity with satellite images.
The first and most obvious of the differences between the view from space and our view of planet Earth is in perspective. The spacecraft imaging devices look straight down at the surface from a great height with monoscopic vision. They use only one eye, so to speak. This perspective is in sharp contrast to our view of Earth that is oblique, from a very low level and in stereo. This oblique, stereoscopic view is processed by our brains to provide (relative) distance and context information. This is how we judge distance and height. Using stereo pairs of aerial photographs, this human capability has proved remarkably useful and accurate in assessing apparent height. Over the last 50 years, much of the topography of the land surface of the Earth has been laboriously mapped in this way.
The second important difference between the view from space and our view of planet Earth is in spatial resolution. With our imaging system the human eye we capture the world as a dense array of points, each of which is infinitely small. This array is combined by our brain to form an image with considerable detail. In addition, we have used our brains in other ways to enhance the spatial resolution of the eye with microscopes, telescopes and spectacles. In contrast to the array of infinitely small points resolved by the human eye, the spatial resolution of the satellite sensor discussed here resolves the Earth's surface into points that are about 60 x 80 metres in size. This means that all the spatial detail within that 60 x 80 metres area is lost because it is represented only in an averaged way. This smearing or averaging of spatial detail is a new way of thinking about what you are seeing. With experience, this concept will becomes less disquieting.
The third important difference between the view from space and our view of planet Earth, is its spectral characteristics. Satellite imaging systems are designed to record the world in spectral wavebands, or colours, that are very different from our visual system. Therefore an obvious and correct conclusion to draw is that the colours of satellite images must be 'false'. The colours in the satellite images used in this book are not 'true' in human terms even though they are natural.
The fourth, last, and most un-human characteristic of satellite data is that it is acquired by machines: machines that are designed to acquire data objectively and repetitively. These two adjectives, objectively and repetitively, say it all because they describe the two characteristics of satellite image data that contribute the most to assessing or monitoring the environment.
It is the two characteristics of objectivity and repetition that have encouraged me to devote my research career to encouraging the use of satellite data for monitoring the Australian environment and for renewable resource management.
Of the four characteristics of satellite data discussed, by far the most critical to us are the spectral and spatial properties. It is these two features that underpin the obvious questions asked of the wildfire image above: what size is this fire or what is the scale of the image and why is the smoke grey-white while the land surface is yellow-red?
The first of these two questions is the easiest to answer and it will help introduce the more difficult answer to the second question.
The size of the fire and the scale of the wildfire can be calculated by looking at the whole satellite image, of which the fire area was but a small part. An entire image from the MSS imaging device of the Landsat satellite series measures 185 x 185 kilometres and is therefore about 34, 000 square kilometres in area.
The entire image of the Dry Season wildfire in Arnhem Land, NT, as recorded by Landsat Multi-Spectral Scanner (MSS) at 0930 hours Australian Eastern Standard Time (AEST) on September 2, 1986. The centre of this scene is 13° 2' South, 133° 29' East.
Details: The image identification is WRS104-069, ID = 58267-00414, 02/09/86. The image size is 185 km by 185 km, or ~34,225 km2.
Obviously this wildfire was not the only one burning in this part of the Northern Territory on this date. Several smaller fires are visible, and you can easily detect the footprints of others. A close examination of the surface shows it criss-crossed with enormous cracks or fissures. Each of these fissures is filled with a crimson ribbon of rainforest. The view from space shows us the bones of the country. This image gives us an insight into the geological forces that shape the Arnhem Land plateau. It is not difficult to speculate as to the how and why of these huge cracks. Also, if I tell you that the eastern edge of this image covers part of Kakadu National Park, then you will be able to interpret the nature and location of one of this Park's most spectacular attractions.
Even though the image was acquired in September, The Dry season in the NT, it is not unreasonable to expect that at least part of the landcover, the vegetation would be green. If the colour of the smoke plumes in these two images is white and that of the fire footprint is black, both of which are natural colours, why is the landcover yellow-red and not green? The colours that we see in this image are not the same as we experience on the ground.
Very simply, the answer to all these questions is the same. The colours of this image are false.
There are good reasons why it is necessary to use false colours in space images. It is necessary now to include a little more detail of the process by which space images are acquired and displayed. The principal issues are the spatial and spectral representations of the surface of the Earth.
The sensors onboard satellites image the Earth's surface by optically capturing the sunlight reflected from the surface through the atmosphere. This imaging process is not the same as a camera that records all the scene (field of view) at the instant the shutter is activated. Rather, the imaging device samples the reflected sunlight continuously by scanning across the direction (path) of the satellite's movement.
As the satellite moves in its orbit, so the imaging device scans the land surface 900 km or so below. Each scanning unit is a pixel (picture element). In the Landsat satellite series this scanning is effected by a moving mirror. The overall process is not very different from that used in digital 35 mm cameras. It is just that the satellite imaging systems operating at the height they do must be uncommonly accurate and precise.
All satellite imaging devices do not record the reflected solar radiation from the Earth's surface in all wavelengths (colours). Rather, they are designed to filter the incoming light into a limited number of groups of wavelengths (wavebands) to reduce the amount of data that have to be recorded to that which is of most interest. For example the Landsat MSS instrument that provided almost all the images in this book records the reflected light from the surface in just four wavebands.
The four channels of data created by the MSS device are mysteriously named MSS#4 - MSS#7, rather than MSS#1 - MSS#4, because there once was an associated imaging device called the Return Beam Vidicon (RBV) that generated the absent channels #1 - #3.
The four wavebands are described by the range of wavelengths of light that the filtering devices let through. As an example, the MSS#4 waveband includes all wavelengths of light from 500 - 600 nanometres (nm), where a nanometre is one billionth (0.000000001) of a metre.
The wildfire scene that we examined earlier is presented here as four black & white images: one for each of the four MSS wavebands. The grey tone in each image is a linear measure of the reflectance of the landscape: the brighter the tone, the greater is the reflectance (brightness) of the landscape in that waveband. The information in these four images is contained in the differences between images. For example, check out how the brightness of the rainforests within the fissures changes between MSS wavebands.
MSS Band 4 GREEN (500-600 nm)
MSS Band 5 RED (600-700 nm)
MSS Band 6 Near Infrared (700-800 nm)
MSS Band 7 Near Infrared (800-1100 nm)
I have associated a colour with each of the first two MSS wavebands (MSS#4 = green, MSS#5 = red), but it is important to remember that colour is a psycho-physical phenomenon. It is a human definition and a concept derived from the structure and functioning of our eyes combined with the processing of this information by our brains.
Colour is not an objective physical phenomenon.
The wavelengths of radiation perceived by human eyes (400 - 700 nm) are collectively called visible light. Human eyes have their peak sensitivities in the yellow-green, this is most probably a consequence of the evolutionary history of hominids that began in canopies of rainforest trees.
The important point is that visible light is not in any physical way different from those wavelengths that we cannot see, eg. the near-infra red channels MSS#6 and MSS#7, nor from the thermal wavelengths that we can feel as heat but are also unable to see.
Why are wavelengths that we cannot see imaged by satellites?
There are two explanations. The first is related to the scattering properties of the atmosphere that is always in between the spacecraft and the earth's surface. The scattering of light by the atmosphere generates fuzzy images of little value. The smaller wavelengths, MSS#4 and MSS#5, are more affected by this atmospheric scattering than the longer ones, MSS#6 & MSS#7. Thus one explanation for recording reflected radiation outside the visible wavebands is the provision of images least affected by atmospheric haze.
The second and more important explanation why satellite imaging devices record wavelengths to which the human eye is not sensitive is that outside our visible world some really interesting things occur. Various parts of the landscape are very reflective (bright) or poorly reflective (dull) in different wavelengths. These differences in the reflectance characteristics of soil, plants, water, shadows, etc, in the non-visible wavebands are a little difficult to comprehend at first contact because they are outside our experience.
It is helpful to begin with an example: vegetation.
Actively growing vigorous plant leaves are quite familiar in the visible: they appear green because they are very reflective (bright) in the green wavelengths (ie. MSS#4) and poorly reflective (dull) in the red (MSS#5). Outside our visible world, instruments tell us that these plant leaves are very bright in the near infrared wavebands, MSS#6 & MSS#7. As leaves become older and less vigorous, they lose their high reflectivity in the near infrared.
On the basis of this information on the spectral attributes of vegetation in the MSS wavebands, we could design a test for green vegetation. If a target was dull (poorly reflective) in the red (MSS#5) as well as being bright (highly reflective) in the near infrared (MSS#6 or MSS#7), then that target is actively growing vegetation.
This test for vegetation is straightforward reasoning from what we know about reflectance.
Note that the test did not use the one characteristic that we humans use to decide whether vegetation is green or not, its colour. The green waveband (MSS#4) was not used at all.
The high reflectance in the near infrared of growing vegetation is not apparent to the human eye in any way; this has been exploited in military activities involving camouflage.
Camouflage is non-living plant material, eg. canvas, coloured and shaped to resemble foliage, used to cover equipment from hostile attack. Well-made camouflage can mislead the trained but unaided eye. So the unaided human eye was assisted by a special photographic film with dyes sensitive in the near-infrared wavelengths. The photographic image collected on this camouflage-detection film clearly showed real vegetation as bright and false vegetation as dull.
The near infrared wavebands of spacecraft imaging devices are designed for the same reasons but for a very different purpose. The non-colour wavebands are included to provide information about the condition or vigour of vegetation. The spectral characteristics of landscape components other than vegetation will be discussed as they occur in later images.
There is one remaining step to answer the most puzzling question that flowed from our wildfire image: given that the colours are false, why is the land surface yellow-red?
This step completes the loop from sunlight reflected from the surface of the Earth, its electronic capture by the imaging device onboard the satellite to the coloured prints in this book.
The MSS imaging device electronically records a scene as an array of numbers which represent the brightness values of the Earth's surface in each of the four wavebands for every pixel. This array, about 28 000 000, was relayed to a ground receiving station. Here these numbers were archived on magnetic tape, carefully preserving their spatial pattern.
A useful analogy at this point is to think of an image as being a mosaic or jigsaw puzzle of very tiny identically shaped tiles. The brightness value of each tile, in each of four colours, is recorded as a digital number (DN) between 0 and 255. The image mosaic will only retain its information content if the spatial pattern of the four brightness numbers is preserved absolutely, that is the spatial correspondence between the brightness values for each of the four wavebands is retained for every one of the 28 000 000 pixels.
The next procedure is to take the array of numbers stored on magnetic tape which has no meaning for the human eye, and faithfully recreate an image. This image must be an exact spatial and spectral copy of the original scene. The human eye and brain are superbly efficient at analysing the spatial patterns of colours that make up an image. In a few ways, the human capacity for image analysis still cannot be matched by a computer.
If the spatial pattern of the pixels can be preserved and displayed, what options are there to display the spectral information?
The most common method of colouring the image is to combine the colours in an additive way. That is, the primary colours of blue, green and red are combined on a black background to generate the colour spectrum. The additive mixing process is controlled by the recorded intensities, the DN values, of the reflected light in three channels of the satellite image. In recreating an image, the DNs control the intensities of spots of blue, green and red light focussed onto a photographic film in a hardcopy output device.
With four original channels of data in the original MSS image and only three primary colours available, a choice in the assignment of colour to waveband must be made.
For Landsat MSS, the most frequent assignment of colours to MSS waveband intensities is blue to MSS#4, green to MSS#5 and red to MSS#7. The data in MSS#6 are thus ignored. For any one of the millions of pixels in the output image, the recorded brightness (DN) of the land surface in MSS#4 (green) controls the amount of blue colour added. The DN for MSS#5 (red) controls the amount of green added to each pixel in the image and the MSS#7 (near infrared) DN controls the amount of red.
Recreating an image by using colour additive mixing and this colour assignment results in a traditional False Colour Composite (FCC) image.
It is not the only way of recreating an MSS image. It is possible to use a subtractive rather than additive colour mixing process in which the yellow, magenta and cyan primaries subtract from a white background to generate the colour spectrum. In addition, the colour assignment between MSS channels is variable because there are 22 different possible ways of combining three channels from four. All of the 22 combinations will produce a false colour composite image. The choice finally depends on the gamut of colours the interpreter prefers.
Finally, let's complete the loop that we challenged ourselves with after examining the wildfire image. Why, in that image is the land surface yellow-red?
The explanation now seems simple. If we had an area of landscape in the NT on the morning of September 2, 1986 that had a cover of green vigorous vegetation, it would have been recorded as bright in the near infrared wavebands (MSS#6 and MSS#7).
Therefore when we come to translate this digital record back into hardcopy and colours, green, vigorous vegetation will appear red in the output FCC image.
The Landsat image archive for Australia began in late 1972 with the launch of Landsat-1. As I write, Landsat-5 is the operational satellite that is still steadily and systematically observing the Earth using both the MSS and the Thematic Mapper (TM) imaging systems.
The MSS imaging system will eventually be replaced by the Thematic Mapper (TM) imaging system that has higher spatial resolution, 30 x 30 metre pixel size rather than the 60 x 80 metres of MSS, and higher spectral resolution, operating with seven spectral wavebands rather than the four of the MSS system. The MSS system has acquired data that now form an enormous image archive of the Earth. The total number of images is not precisely known, but it is more than 2 million.
The principal imaging device aboard Landsat 1-5 has been the MSS. With the launch of Landsat-4 in 1983, the Thematic Mapper (TM) was added to the payload with higher spectral and spatial resolution but with the same swathe width and revisit time.
In summary: Satellite images are unique and extremely powerful in conveying information about the environment. They are unusual and unfamiliar because the view that they provide is new and challenging. We can use that view to provide an objective assessment of the state of environment twenty years ago and as it is today. Satellite data are a new truth and a new history. We can use it to detect the changes that have taken place over that time span and to determine whether those changes were for the better, benign, or for the worse. We can objectively determine the rates of change, the spatial location of change and, most importantly, who was responsible for that change. The management of the Earth has never had such a powerful tool at its disposal, yet that tool resides 900 kilometres out in space.
Calder, N., 1991., 'Spaceship Earth'. Penguin Books Ltd, London.
Drury, S.A., 1990., 'A guide to remote sensing: Interpreting images of the Earth'. Oxford Science Publications, Oxford.
Harrison, B.A. and Jupp, D.L.B., 1989., 'Introduction to remotely sensed data'. CSIRO, Australia.
Sheffield, C., 1981., 'Earth Watch: A survey of the world from space'. Hutchinson Group (Australia) Pty. Ltd., Richmond, Victoria.
