The SPOT Satellite: A Case Study in Africa; Further Applications to Vegetation Mapping of Crops and Rangelands - Lecture Note - Completely Remote Sensing tutorial, GPS, and GIS -
The SPOT Satellite: A Case Study in Africa; Further Applications to Vegetation Mapping of Crops and Rangelands

We interject the following synopsis of another operational satellite which, like Landsat, can provide very sharp and timely images well suited to agricultural analysis. We will then use SPOT to conduct a study of vegetation in Africa.

SPOT is an operational series of Earth-observing satellites, sponsored initially by the French government and managed by the French Space Agency, which, since the first launch in 1986, offers commercial multispectral data to users worldwide. From its 831 km (516 mi), sun-synchronous (10:30 A.M. equatorial crossing), near-polar orbit, SPOT revisits the same 117 km-wide (73 mi) ground tracks every 26 days, when its sensors look vertically downward (at nadir). However, operations can move its mirror sideways in both directions up to 27 degrees off-nadir, enabling it to view a 950 km-wide (590) corridor along its ground track. With careful planning, and optimal cloud conditions, it can revisit an area within this corridor seven times during a 26 day orbit cycle.

Images of the same area taken from different lateral angles on separate orbital passes can be paired, so that their parallax offsets permit stereoscopic viewing, thus facilitating a 3-D perspective of the area (see Section 11). These images are of great value to topographic analysts, geomorphologists, and others.

The payload sensor on the first SPOTs is the HRV (High Resolution Visible) imager. This name is slightly misleading, because in its multispectral mode, Band 3 operates in the near-IR wavelength over the interval 0.79 - 0.89 µm (Band 1 = 0.50 - 0.59 µm [green] and Band 2 = 0.6 - 0.68 µm [red]). A separate unit in the HRV can also image in the panchromatic mode as a single band encompassing 0.51 - 0.70 µm. Instead of an oscillating scan mirror, the tiltable mirror (set either on or off-nadir) sends light from a scene to a linear array of tiny detectors, known as charged coupled devices (CCDs). (See page I-5a for details on how these devices function.) The line of CCDs receives photons simultaneously and very rapidly discharges in sequence into an electron current stream. This array is then reactivated with photons almost instantly, as the array line moves to the next ground position along the forward track. In this way, the scene is developed line by line. In the panchromatic mode, there are 6,000 detectors in the linear array, each receiving light from a 10 x 10 m ground area, thus providing a resolution of 10 m for an image of 60 km (37 mi) width. The multispectral HRV contains 3000 detectors in the array, providing a ground resolution of 20 m. On each spacecraft there are two such HRV arrays, mounted side-by-side, each viewing its own 60 km swath width but the pair of images sidelapping to give an effective 117 km (73 miles) wide coverage.

Later SPOT satellites have improved resolution. They cover smaller scene areas. Many SPOT images are near ideal for certain applications such as viewing a city in context with its surroundings. As an example, look at this SPOT image of Marseille, France:

Marseille, France.
SPOT has a large number of image on the Web, under the categories of Nature and Cities, that can be accessed on SPOT gallery.

Kenya and the African Rift Valley

Now that we've described SPOT, lets look at one of its images and see what you can spot (couldn't pass up that one!). In Section 2 you have earlier examined a SPOT scene of the Great African Rift valley which we will repeat here for context.

SPOT scene of the African Rift in Kenya.

To put the SPOT image in context, we need first to say something about the Rift Valley. This rift zone, one of the largest in the world, is astride a regional uplift, where part of the eastern African continent is pulling away (an incipient divergent boundary in the plate tectonics model). Within the rift, parts of the crust are dropping down along normal faults in a series of one or more steps as shown in the figure below. If normal faults are paired on opposite sides of a valley, this is evidence of a strutural graben, shown diagrammatically here:.

Cross-sectional sketch of production of grabens (downdropped blocks) and horsts (upthrown blocks) by rifting along normal faults; typical of the African Rift Valley structural geology.

Some faults are young enough to remain relatively uneroded, so that they present a steep, often abrupt, face (called a scarp), separating the upthrown (upland) block from the downthrown (valley) block, as evident in the ground photos below.

 Aerial oblique photo of a small segment of the East African Rift in which vegetation is sparse; the steep walls define the upthrown sides of Normal (type) faults.

 Aerial oblique photo of another part of the East African Rift; here, greater rainfall has modified the cliffs by erosion and the entire area is more heavily vegetated.

Along much of the rift, the land is an arid to semi-desert savannah (top photo, above). In other segments, the scarp is older, and weathered, and may be covered in part with forest or dense shrub, such as in the lower (or right) photo above. Most of the rift zone is underlain by a sequence of basaltic (volcanic) flow units. The next scene is a regional mosaic made from all or parts of nine Landsat MSS images, of much of Kenya that extends over an area about 500 km (311 mile) wide, east-west.

Landsat-2 images (false color) combined in a mosaic covering a portion of the East African Rift in Kenya.
Index map showing main features in Landsat image to the left.

Several of the volcanic mountains, in particular Mt. Kilimanjaro shown at the bottom of the map, are displayed as a nearly horizontal perspective view using combined Landsat-SRTM data:

SRTM-Landsat image of the mountains in the lower part of the index map above.

Match major landmarks in the above mosaic, with those in the index map including Mt. Kenya (in Kenya) and Mt. Kilimanjaro in Tanzania, the latter being over 5,230 m (17,154 ft) high and therefore, snow-capped. Kilimanjaro supports active glaciers, as seen from this International Space Station as photographed by an astronaut:

Kilimanjaro as seen from the ISS.

Now to the specific subscene that we will examine for vegetation and classify for crops. This example is an agricultural and forestry scene, just east of the town of Nakuru in Kenya, about 100 kms (60 mi) northwest of Nairobi, the capital. That town (shown on page 6-13a) lies within the Great African Rift Valley of East Africa (described earlier on page 2-9). The 10 km-wide subscene below was extracted from a full 60 km SPOT image, which was produced as a false color composite. We will analyze this subscene in the next paragraph.

A SPOT three band false color composite of a segment of the East African Rift in Kenya where much of the land is suitable for farming and natural forests.

Whereas details are lacking in the mosaic, the SPOT image shows a plentiful amount of surface information. The edge of the Rift Valley passes from upper center to lower right, as marked by dark red areas that are associated with forest lands. To its right is an area of rather uniform red, without any treelike textural pattern, which are interpreted as grass and shrub lands. The area to the left (west) is occupied mainly by irregular-shaped fallow fields, many of which are about 0.8 km (0.5 mi) long. Some fields appear in this color version as almost white, whereas others are in three dominant shades of blue (darker, medium, light blue). Inspecting the three individual SPOT bands (not reproduced here), the white fields are nearly white in each band and the blue fields are medium gray in band 1, darker in Band 2, and very dark gray in Band 3. The lack of any red areas in these two types of fields confirms that crop growth (mainly of coffee and wheat in this part of Kenya) is nil or too early to appear. Not having been there, we do not have any adequate explanation for this pronounced difference in tone between these fields. The dark tones could represent the prevailing color (also dark) for soils derived from the very dark basalts. But this fails to explain the light blue tones for the other fields. They might be the signature of a grain crop, such as unharvested mature wheat (hence appearing bright straw yellow as seen onsite), thus being very reflective in Band 1, much less so in Band 2 (ground reds), and least reflective in the IR band. Any further speculation without supporting ground truth would be frustrating. Note the pink tones in several areas of the scene including along stream courses and in the narrow stream canyons that are cut into the rift wall slopes. These pink tones may be due to two conditions: the foliage of some trees has only begun to develop new growth in this March 14, 1986 scene and the vegetation there is experiencing some stress due to deforestation as local farmers progressively clear the land.

Some of the above ideas are implicit in the supervised maximum likelihood classification of this scene, in which the classes chosen are: woods, grasses, newclear (new growth and cleared land), ltfield, medfield, dkfield (light, medium and dark fields). The results are quite believable, in that the color patterns tend to be homogeneous and their distributions are logical. The extent of newclear is somewhat greater than we can discern in black & white images. There are some black areas in the classified image. Those black areas that appear scattered are simply unclassified pixels. But several areas have definite shapes and are probably small lakes or ponds ( there are flamingo breeding grounds in this region).

IDRISI-made quasi-supervised classification (the classes were arbitrarily set up by photo-interpretation rather than from field evidence) of the above SPOT scene of an area in Kenya.

Quite by serendipity, during the preparation of this caption, we discovered that the IDRISI program uses two special theme maps of the Nakuru region as part of its training exercises. With permission, we repeat these here. On the left is an image map of elevations, in feet, showing the Rift Valley to be as low as 5,690 ft (1,734 m), at a point within the blue pattern, and the western uplands to reach to greater than 9,850 ft (3,002 m) in yellow. The SPOT scene fits somewhere in the upper right. The map on the right is another map, which plots the annual rainfall in millimeters (green is wettest, then red, yellow, and white progressively driest). (Note that in this equatorial region of East Africa there are two rainy seasons: winter and summer.) Try drawing your own relationships to the SPOT scene from these maps.

Elevation map of the Nakuru region.
Mean Annual Rainfall map of the Nakuru region.

SPOT is effective at one mode of change detection using multitemporal imagery, applying to the same scene from two dates a point by point merging of pixels and then a classification algorithm, to determine in the case study shown below the variations in productivity of maize in Zimbabwe:

Variations in productivity of maize

While in Africa, make note of these unusual scenes, in the desert of Sudan drained here by the Upper Nile. Water irrigated from this river has turned a once barren area into a vast array of farms (mostly cotton and millet). These are elongated in the style of the French Long Lot patterns (see also the wider scope image of this farming development on page 6-13).

Details of Sudan irrigation farms; similar patterns occur in parts of Louisiana in the U.S.
More farming patterns in the Sudan; SPOT image.

Here is a stellar example of classification of crops, this time using a Landsat TM data set for July 7, 1993 that covers numerous farms along coastal Holland.

Classification of TM data into crop types on farms along the coast of Holland.

Unfortunately, the legend writing is obscured, but from top to bottom the crops are: Potato; Sugar Beets; Wheat; Grass; Maize; Rapeseed; Barley; and Lucerne (a type of alfalfa), and forest. Although colorful and rather convincing in its patterns, the accuracy of identification was only 56% - not uncommon when certain conditions needed to achieve higher accuracies are not met. We will encounter this scene again in page 13-4c during the discussion of multitemporal monitoring of crops.

As asserted on the previous page, the spectral response of vegetative matter does not show much difference between various species or types. The spatial aspect of crops, for example, in plant shapes, leaf arrangement, stage of growth, etc. are often the best distinguishing factors. However, as was also stated, detailed spectral curves show subtle variations that can be diagnostic - hence, hyperspectral remote sensing is commonly the best tool for crop identification, as was demonstrated using the Hyperion sensor. We underwrite that postulate with this look at the use of AVIRIS to identify and differentiate crops in the San Juan Valley near Summerville, Colorado. First, examine this set of spectral curves.

Spectral curves (offset) for 5 crops types at the Summerville, CO test plot.

Using those curves, the fields (many circular for adaptability to pivot irrigation; see next page) were identified as to crop type (see caption) using supervised classification, giving this result (>80% accuracy):

Classified image of fields near Summerville, CO; the legend includes Bare Field; Barley; Pasture; and two types of Potatoes.

These plots are actual reflectances, rather than offset (as on page 3-1). Certain absorption bands and the height of a spectral curve for a given crop type will provide evidence of a deficiency of plant moisture that leads to a state of crop stress. Some of the individual field crops in the Summerville scene had insufficient moisture (thus, are stressed) at the time AVIRIS sensed them during an aerial flyby. This map shows this condition:

Variation in levels of internal water content, indicating some moisture stress, in the Summerville crops.

Many legum crops such as wheat and barley resemble grasses. Both natural and planted grasses are commonly the prime crop in rangelands, where they serve as feed for grazing animals. Landsat TM data can identify and map grasslands, as is demonstrated in this next example, reported by Dr. Paul Tueller of the University of Nevada-Reno, a world expert in this application. The TM subscene occur near the Organ Mountains in the piedmont covered by the open rangelands that occur within Fort Bliss, Texas (where NMS served his military career). First, the subscene, then the classification:

False color TM subscene of an area within the Fort Bliss, TX military preserve, consisting of rocky soil, a ridge, and varying amounts of grassland cover.
Classification of grasslands.

This classification shows six categories of vegetation, plus a modicum of bare land, a feat of identification that is impressive indeed.