Vegetation Application: General Principles For Recognizing Vegetation - Lecture Note - Completely Remote Sensing tutorial, GPS, and GIS - facegis.com
Vegetation Application: General Principles For Recognizing Vegetation

Planet Earth is distinguished from other Solar System planets by two major categories: Oceans and Land Vegetation. The oceans cover ~70% of the Earth's surface; land comprises 30%. On the land itself, the first order categories break down as follows: Trees = 30%; Grasses = 30%; Snow and Ice = 15%; Bare Rock = 18%; Sand and Desert Rock = 7%.

We have already seen in previous Sections and in the Overview that in false color imagery the remote sensing signature of vegetation is a bright red. The landscape shown in this first image could almost be on Mars except for the presence of this bright red sign of vegetation. This is the Ouargla Oasis in the Sahara Desert of southern Algeria, a concentration of trees and plants where groundwater reaches the surface:

The Ouargla Oasis.

On Earth, the amount of vegetation within the seas is huge and important in the food chain. But for people the land provides most of the vegetation within the human diet. The primary categories of land vegetation (biomes) and their proportions is shown in this pie chart:

The distribution of the main types of vegetation biomes, worldwide.

These biomes are defined in part by the temperature and precipitation controls that differentiate them:

Differences in biomes as determined by temperature and precipitation variations.

Global maps of vegetation biomes on the continents show this general distribution:

Worldwide distribution of vegetation biomes.

A fair number of global vegetation maps have been published. These usually show slight to moderate differences, depending in part with the types and numbers of classes established in the classification. Compare the map below with the one above:

World vegetation map.

There is a notable correlation between vegetation classes and climate. Once again, compare the climate map below with the two vegetation maps.

Broad classification of climate types, with indications of their relation to vegetation.

Remote sensing has proven a powerful "tool" for assessing the identity, characteristics, and growth potential of most kinds of vegetative matter at several levels (from biomes to individual plants). Vegetation behavior depends on the nature of the vegetation itself, its interactions with solar radiation and other climate factors, and the availability of chemical nutrients and water within the host medium (usually soil, or water in marine environments). A common measure of the status of a given plant, such as a crop used for human consumption, is its potential productivity (one such parameter has units of bushels/acre or tons/hectare, or similar units). Productivity is sensitive to amounts of incoming solar radiation and precipitation (both influence the regional climate), soil chemistry, water retention factors, and plant type. Examine the diagram below to see how these interact, keeping in mind that various remote sensing systems (e.g., meteorological or earth-observing satellites) can provide inputs to productivity estimation:

Flow chart showing inputs and interconnections that need to be quantified to estimate vegetation productivity.

Because many remote sensing devices operate in the green, red, and near infrared regions of the electromagnetic spectrum, they can discriminate radiation absorption and reflectance properties of vegetation. One special characteristic of vegetation is that leaves, a common manifestation, are partly transparent allowing some of the radiation to pass through (often reaching the ground, which reflects its own signature). The general behavior of incoming and outgoing radiation that acts on a leaf is shown here:

Partition of incoming solar irradiation interacting with a leaf; some is transmitted through the leaf, some is reflected, and some is absorbed, leading to re-radiation as heat at longer wavelengths.

Now, consider this diagram which traces the influence of green leafy material on incoming and reflected radiation.

The Interaction between light and the components of a leaf.

Absorption centered at about 0.65 Ám (visible red) is controlled by chlorophyll pigment in green-leaf chloroplasts that reside in the outer or Palisade leaf. Absorption occurs to a similar extent in the blue. With these colors thus removed from white light, the predominant but diminished reflectance of visible wavelengths is concentrated in the green. Thus, most vegetation has a green-leafy color. There is also strong reflectance between 0.7 and 1.0 Ám (near IR) in the spongy mesophyll cells located in the interior or back of a leaf, within which light reflects mainly at cell wall/air space interfaces, much of which emerges as strong reflection rays. The intensity of this reflectance is commonly greater (higher percentage) than from most inorganic materials, so vegetation appears bright in the near-IR wavelengths (which, fortunately, is beyond the response of mammalian eyes). These properties of vegetation account for their tonal signatures on multispectral images: darker tones in the blue and, especially red, bands, somewhat lighter in the green band, and notably light in the near-IR bands (maximum in Landsat's Multispectral Scanner Bands 6 and 7 and Thematic Mapper Band 4 and SPOT's Band 3).

Identifying vegetation in remote-sensing images depends on several plant characteristics. For instance, in general, deciduous leaves tend to be more reflective than evergreen needles. Thus, in infrared color composites, the red colors associated with those bands in the 0.7 - 1.1 Ám interval are normally richer in hue and brighter from tree leaves than from pine needles.

These spectral variations facilitate fairly precise detecting, identifying and monitoring of vegetation on land surfaces and, in some instances, within the oceans and other water bodies. Thus, we can continually assess changes in forests, grasslands and range, shrublands, crops and orchards, and marine plankton, often at quantitative levels. Because vegetation is the dominant component in most ecosystems, we can use remote sensing from air and space to routinely gather valuable information helpful in characterizing and managing of these organic systems.

The ability to distinguish different types of vegetation was brought home to the writer (NMS) through a simple study using a densitometer to examine multispectral images of a strip of agricultural land near the Choptank River in the eastern shore of Maryland. These images were part of an experiment by my first "boss", Dr. Warren Hovis, at Goddard. He had built a multispectral sensor to fly on an aircraft that would simulate images made by the same four bands on the ERTS-1 (Landsat-1) Multispectral Scanner (MSS). Here are the images:

Four MSS-equivalent bands images of fields in the Choptank River test site; the numbered lines in the second image refer to the fields whose gray levels were determined with a densitometer.

The relative gray levels are plotted as a four band histogram for each of the numbered features in the above image. It should be evident that there are real differences in these band signatures among the vegetation and other features present; thus Mixed Hardwoods have different relative "brightness" patterns from Soybeans, from Old Hay, etc..

Histogram signatures of the numbered features in the above set of images.

This discrimination capability implies that one of the most successful applications of multispectral space imagery is monitoring the state of the world's agricultural production. This application includes identifying and differentiating most of the major crop types: wheat, barley, millet, oats, corn, soybeans, rice, and others.

This capability was convincingly demonstrated by an early ERTS-1 classification of several crop types being grown in Holt County, Nebraska. This pair of image subsets, obtained just weeks after launch, indicates what crops were successfully differentiated; the lower image shows the improvement in distinguishing these types by using data from two different dates of image acquisition:

Classification of crop types discriminated using MSS multiband data for fields in Holt Co, NE.

This is a good point in the discussion to introduce the appearance of large area croplands as they are seen in Landsat images. We illustrate with imagery that covers the two major crop growing areas of the United States.

The first scene is part of the Great or Central Valley of California, specifically the San Joaquin Valley. Agricultural here is primarily associated with such cash crops as barley, alfalfa, sugar beets, beans, tomatoes, cotton, grapes, and peach and walnut trees. In July of 1972 most of these fields are nearing full growth. Irrigation from the Sierra Nevada, whose foothills are in the upper right, compensates for the sparsity or rain in summer months (temperatures can be near 100░ F). The eastern Coast Ranges appear at the lower left. The yellow-brown and blue areas flanking the Valley crops are grasslands and chapparal best suited for cattle grazing. The blue areas within the croplands (near the top) are the cities of Stockton and Modesto.

The Great Valley of California, one of the principal vegetable-producing regions of the U.S.

The second Landsat image is in the Wheat Belt of the Great Plains. The image below is of western Kansas in late August. Most of the scene consists of small farms, many of section size (1 square mile). The principal crop is winter wheat which is normally harvested by June. Spring wheat is then planted, along with sorghum, barley, and alfalfa. This scene is transitional, with nearly all of the right side being heavily planted, but the left side (the High Plains, at higher elevations) contains some unplanted farms and cropfree land, some used for grazing.

The central Great Plains; most of the reddish color associates with growing crops and much of he gray-blue is fields not supporting crops by late August.

Still another example of winter wheat in early growth is this scene in southwestern Australia, east of Perth. Some of the wheat fields are quite large - 5 km (3 miles) or more on a side. The prevailing color is tan but with a faint red cast, implying initial growth. There is a sharp line dividing many fields from the mallee scrub (dark brown) growing on soils derived from Precambrian rocks. This line marks an electrified rabbit fence, keeping these "pests" from nibbling on the wheat and other crops being grown.

Large fields in south Astralia, growing mostly wheat; Landsat-1 MSS image.

Many factors combine to cause small to large differences in spectral signatures for the varieties of crops cultivated by man. Generally, we must determine the signature for each crop in a region from representative samples at specific times. However, some crop types have quite similar spectral responses at equivalent growth stages. The differences between crop (plant) types can be fairly small in the Near-Infrared, as shown in these spectral signatures (in which other variables such as soil type, ground moisture, etc. are in effect held constant).

Spectral plots for several plant species; the curves have been offset vertically to allow each to be seen without overlap; actually, the differences due to the spectral response of the vegetative material are small.

The shape of these curves is almost identical when each crop type is compared with the others. The big difference is in the percent reflectance. The similarity in shape is explained by the fact, discussed earlier on this page, that most vegetation matter has the same basic cell structure and similar content of chlorophyll. Yet remote sensing is reasonably effective at distinguishing and identifying different crop types. Why is that possible? These next two illustrations afford several clues:

Fields of different crops.
Some visual differences among multiple crop types.

3-1: Drawing on your experience and common sense, make (or think) a list of the factors that will affect the spectral signatures of field crops, and thus help to separate crop types. ANSWER

Read the answer to this question - it is important. The list is incomplete, but the main factors are discussed. But with so many variables involved, it is difficult to claim that each crop has a specific spectral signature. This means that, in order to identify the several crops usually present in agricultural terrain in any particular area, the most efficient course is to establish training sites, as was discussed on the pages dealing with supervised classification in Section 1.

As we learned in the Introduction Section, spectral characteristics are one means of identifying and classifying features in a scene. We will see how reliable this is by itself as this Section unfolds. We also learned in the Introduction that shape and pattern recognition are valuable inputs in determining what a feature is. The geometric shape of a field of crops sometimes is helpful in determining the actual crop itself. But field shapes tend to vary both within regions of large countries like the U.S. and in different parts of the world. This variation is evident in the illustration below (read the caption to find out which country goes with a particular panel):

Various crop field patterns: Upper left = Minnesota; Upper center = Kansas; Upper right = Northwest Germany; Lower left = Bolivia; Lower center = Thailand; Lower right = Brazil; these are ASTER subscenes.

Through remote sensing it is possible to quantify on a global scale the total acreage dedicated to these and other crops at any time. Of particular import is the utility of space observations to accurately estimate (goal: best case 90%) the expected yields (production in bushels or other units) of each crop, locally, regionally or globally. We do this by first computing the areas dedicated to each crop, and then incorporating reliable yield assessments per unit area, which agronomists can measure at representative ground-truth sites (in the U.S., county farm agents obtain routinely from the farmers themselves). Reliability is enhanced by using the repeat coverage of the croplands afforded by the cyclical satellite orbits assuming, of course, cloud cover is sparse enough to foster several good looks during the growing season. Usually, the yield estimates obtained from satellite data are more comprehensive and earlier (often by weeks) than determined conventionally as harvesting approaches. Information about soil moisture content, often critical to good production, can be qualitatively (and under favorable conditions, quantitatively) appraised with certain satellite observations; that information can be used to warn farmers of any impending drought conditions.

Under suitable circumstances, it is feasible to detect crop stress generally from moisture deficiency or disease and pests, and sometimes suggest treatment before the farmers become aware of problems. Stress is indicated by a progressive decrease in Near-IR reflectance accompanied by a reversal in Short-Wave IR reflectance, as shown in this general diagram:

General pattern of change in the spectral curve of a species as the plant experiences stress resulting from moisture deficiency.

This effect is evidenced quantitatively in this set of field spectral measurements of leaves taken from soybean plants as these underwent increasing stress that causes loss of water and breakdown of cell walls.

Stress effects in soybeans.

For the soybeans, the major change with progressive stress is the decrease in infrared reflectances. In the visible, the change may be limited to color modification (loss of greeness), as indicated in this sugar beets example, in which the leaves have browned:

The small differences (maximum in the red) in the spectral signatures of normal and stressed beets.

Differences in vegetation vigor, resulting from variable stress, are especially evident when Near Infrared imagery or data are used. In this aerial photo made with Color IR film shows a woodlands with healthy trees in red, and "sick" (stressed) vegetation in yellow-white (the red no longer dominates):

Color IR photo showing healthy vegetation in red and stressed vegetation in blues and light yellows.

For identifying crops, two important parameters are the size and shape of the crop type. For example, soybeans have spread out leaf clumps and corn has tall stalks with long, narrow leaves and thin, tassle-topped stems. Wheat (in the cereal grass family) has long thin central stems with a few small, bent leaves on short branches, all topped by a head containing the kernels from which flour is made. Other considerations are the surface area of individual leaves, the plant height and amount of shadow it casts, and the spacing or other planting geometries of row crops (the normal arrangement of legumes, feed crops, and fruit orchards). The stage of growth (degree of crop maturity) is also a factor. For example during its development wheat passes through several distinct steps such as developing its kernel-bearing head and changing from shades of green to golden-brown (see below).

Another related parameter is Leaf Area Index (LAI), defined as the ratio of one-half the total area of leaves (the other half is the underside) in vegetation to the total surface area containing that vegetation. If all the leaves were removed from a tree canopy and laid on the ground, their combined areas relative to the ground area projected beneath the canopy would be some number greater than 1 but usually less than 10. As a tree, for example, fully leaves, it will produce some LAI value that is dependent on leaf size and shape, the number of limbs, and other factors. The LAI is related to the the total biomass (amount of vegetative matter [live and dead] per unit area, usually measured in units of tons or kilograms per hectare [2.47 acres]) in the plant and to various measures of Vegetation Index (see below). Estimates of biomass can be carried out with variable reliability using remote sensing inputs, provided there is good supporting field data and the quantitative (mathematical) models are efficient. Both LAI and NDVI (page 3-4) are used in the calculations.

Satellite remote sensing is an excellent means of determining LAI on a regional or subcontinental scale. Here is an LAI map of northern South America and southern Central America obtained using Landsat data. The Amazon rainforest (see page 3-5) and the tropical forests on the eastern slopes of Costa Rica and Panama are well-defined by the LAI distribution (they diminish rapidly, in part because of rainfall patterns):

LAI map of parts of Central and South America.

Studies of LAI for the Amazon vegetation led to an unexpected discovery. Examine this map:

Seasonal variations in LAI in the Amazon Basin.

The largest change in the "greeness" of the tree leaves in the Amazon occurred, not as anticipated in the wet season, but in the dry season. The explanation: water is stored in the near surface vadose zone (part of upper soil in which groundwater percolates downward) and below the top of the water table. That water is tapped by the plants during the dry period so that leaves are able to continue growth.

In principal, actual LAI must be determined on site directly by stripping off all leaves, but in practice it can be estimated by statistical sampling or by measuring some property such as reflectance. Thus, remote sensing can determine an LAI estimate if the reflectances are matched with appropriate field truth. For remotely sensed crops, LAI is influenced by the amount of reflecting soil between plant (thus looking straight down will see both corn and soil but at maturity a cornfield seems closely spaced when viewed from the side). For the spectral signatures shown below, the Near IR reflectances will increase with LAI.

Changes in the spectral response curve of soybeans with increasing LAI (from early growth stage to maturity).

This change in appearance and extent of surface area coverage over time is the hallmark of vegetation as compared with most other categories of ground features (especially those not weather-related). Crops in particular show strong changes in the course of a growing season, as illustrated here for these three stages - bare soil in field (A); full growth (B); fall senescence (C), seen in a false color rendition:

Stages of crop development (concentrate on the field in the center).

3-2: How would non-growing or dead vegetation (such as crops in senescence) be detected by Landsat? ANSWER

The study of vegetation dynamics in terms of climatically-driven changes that take place over a growing season is called phenology. A good example of how repetitive satellite observations can provide updated information on the phenological history of natural vegetation and crops during a single cycle of Spring-Summer growth is this sequence of AVHRR images of the Amu-Dar'ja Delta just south of the Aral Sea in Ujbekistan (south-central Asia).The amount of vegetation present in the delta (a major farming district for this region) is expressed as the NDVI, an index defined on page 3-4. The Aral Sea - a large inland lake - is now rapidly drying up (see page page 14-15).

NDVI classification of an AVHRR image of the Aral Sea region.

More generally, seasonal change appears each year with the "greening" that comes with the advent of Spring into Summer as both trees and grasses commence their annual growth. The leafing of trees in particular results in whole regions becoming dominated by active vegetation that is evident when rendered in a multispectral image in green tones. The MODIS sensor on Terra has several vegetation-sensitive bands used to calculate a variation of the NDVI called the Enhanced Vegetation Index (EVI). This trio of images (dates in the caption) shows the spread of growing natural vegetation across the U.S.

Phenological changes across the U.S.of tree leafing and grass growth, shown in enlarging greens, starting with the limited extent of active vegetation in January 2001 (top), then the reawakening of growth in the Southeast during March-April (middle), and the expansion of leaf cover in May-June (bottom); the white areas are snow cover; an image mosaic made from wide area multiple strips using ASTER imagery from the Terra spacecraft.

During the first stages of growth in the Spring season in the eastern half of the U.S., drastic changes in vegetation signatures will ensue. In the 6-panel figure below, the left two panels (in the top and bottom triads) show a near natural color MISR image of a strip in the central U.S. that includes the western Ouachita Mountains of Oklahoma-Texas on April 1, 2004 and May 3, 2004. In the center pair (top and bottom), LAI values are calculated for both dates - by May the barren trees and grasslands had almost fully leafed. On the right is shown a map that displays the fraction of photosynthetically active radiation for the two dates - photosynthesis, as expected, has neared full term by May

MISR images of a part of mid-continental U.S., with changes described in the above paragraph.

A variant of these ideas is the abnormal greening of an area as a result of an extended period of excessive rainfall. This happened in Texas in June of 2007 when as much as 20 inches of rain fell during the month. This mosaic made from SPOT imagery (see next page) shows that most of the western two-thirds of the state has a far greener vegetation response than normal at this time of year (the blackish area, around the Midland Basin, is the extreme):

Now, to emphasize the variability of the spectral response of crops over time, we show these phenological stages for wheat in this sequential illustration:

A diagram in which the top area shows pictorially the stages of growth of winter wheat, each tied to its appearance on the ground (4 middle panels) and as seen in Landsat imagery (4 lower panels).

Note that, in the Landsat imagery, the wheat fields (particularly the light-blue polygon in the far-left image) show their brightest response in the IR (hence red) during the emergent stage but become less responsive by the ripening stage. The grasses and alfalfa that make up pasture crops mature (redden) much later.

3-3: The display above was made as part of the LACIE program, designed to demonstrate that crop history can be monitored by satellite and that productivity (yield in bushels) and prediction of total yield in a season from regions of major productions can be quantitatively assessed. How were the specific crop types used as training sites identified (determined)? ANSWER

With this survey of the role of several variables in determining crop types, let us look now at one of the most successful classifications reported to date. These are being achieved by hyperspectral sensors such as AVIRIS (page 13-9) and Hyperiorn. The Hyperion hyperspectral sensor on NASA's EO-1 ( (Page Intro-24) has procured multichannel data for the Coleambally test area in Australia. This image, made from 3 narrow channels in the visible-Near IR, shows how the fields of corn, rice, and soybeans changed their reflectances during the (southern hemisphere) growing season: Notice the pronounced differences in crop shapes which is a big factor in producing the reflectance differences (as said above, healthy leaf vegetation generally has a spectral response that does not vary much in percent reflectance from one plant type to others, so that differences in crop shape become the distinguishing factor).

False color images of fields in Australia scanned by the Hyperion sensor on EO-1; note the distinct difference in shape and texture of the 3 crops, which is important in determining the spectral response of each.

The multichannel data from Hyperion were used to plot the observed spectral signatures for the soil and three crops, as shown here (the curves identified in the upper right [the writing is too small to be decipherable on most screens] are, from top to bottom, soil, corn, rice, and soybeans):

Spectral signatures of four ground classes derived from data monitored by the Hyperion sensor on EO-1;

Using a large number of selected individual Hyperion channels, this supervised classification of the four classes in the subscene was generated; this end result is more accurate than is normally achievable with broad band data such as obtained by Landsat:

Classification of corn, rice, soybeans, and bare soil in Australian fields, using EO-1 Hyperion data.

Active microwave sensors, or radar, can use several variables to recognize crop vegetation and even develop a classification of crop types. Here is a SIR-C (Space Shuttle) image of farmland in the Netherlands, taken on April 4, 1994. The false color composite was made with L-band in the HH polarization mode = red; L-band HV = green; and C-band HH = blue (see page 8-5).

SIR-C image of crops in the Netherlands.

An additional image variable is the crop's background, namely the nurturing soil, whose color and other properties can change with the particular soil type, and whose reflectance depends on the amount of moisture it holds. Moisture tends to darken a given soil color; this condition is readily picked up in aircraft imagery as seen in this pair of images:

Color aerial photo of a field with little surface moisture. The same field after a rainstorm.

Often, the distribution of moisture, as soil dries differentially, is variable in an imaged barren field giving rise to a mottled or blotchy appearance. Thermal imagery (Section 9) brings out the differential soil moisture content by virtue of temperature variations. The amount of water in the crop itself also affects the sensed temperature (stressed [water deficient] or diseased crop material is generally warmer). Soil water variations are evident in this image made by an airborne thermal sensor of several fields, where high moisture correlates with blue and drier parts of the fields with reds and yellows:

Thermal scanner image of fields with variable moisture.

A combination of visible, NIR, and thermal bands can pick up both water deficiency and the resulting stress on the crops in the fields. This set of three images was made by a Daedalus instrument flown on an aircraft. In the top image, yellow marks unplanted fields and those in blue and green are growing crops. The center image picks up patterns of water distribution in the crop fields. The bottom image shows levels of stress related in part to insufficient moisture.

Airborne scanner images of several fields, indicating crop locations, water deficiency, and stress.

A passive microwave sensor (page 8-8) also picks up soil moisture. Cooler areas appear dark in images of fields overflown by a microwave sensor - although other factors, such as absence or presence of growing crops (and their types) besides moisture can account for some darker tones:

Passive microwave image obtained during an overflight of

Radar likewise can detect variations in soil moisture in agricultural fields. Below is a C-band airborne SAR image of an experimental station at Maricopa, AZ near Phoenix. The darker fields are those with both higher moisture and growing crops which, in this case, result in less signal returns to the SAR receiver.

Radar C-band image of agricultural plots at the Maricopa Experimental Station, AZ.

A major goal in the EOS program (Section 16) is to produce soil moisture maps on a short-term basis (say, for two weeks running, available almost immediately thereafter). Various sensors on Terra and Aqua can provide data needed to calculate regional soil moisture distribution. Here is a map for the entire United States covering the period July 1-15, 2003:

Soil moisture map of the U.S. for July 1-15, 2003.

This map indicates an abundance of moisture in the northeast and upper Central Lowlands (various areas therein have been affected by one of the wettest seasons in recent decades) and a continuing (for several years) drought in most of the western half of the U.S. (a condition responsible for an abnormally high number of forest fires (see Overview).

Before moving on to some specific examples of vegetation analysis, we append this interesting image of a variant of the topic of Crop Signs - which refers to strange, often unexplained circles and other figures cut into maturing crops (commonly corn fields). In this case, the perpetrator was not some mysterious "alien" source but a patriotic farmer named Fritzler whose farm is outside Greeley, Colorado:

Farmer Fritzler's patriotic contribution to military serving in the Iraq War; IKONOS image.

And since this page is ending with the unusual, here is another "oddity" that defies categorization (which means it is hard to find a proper page to display it). Farming calls to mind crops that are 'vegetation'. In a broader sense, animals are said to be farm-raised. But fish!! Yet aquaculture is a growing industry. For example, catfish are now mostly harvested from ponds on 'farms' that produce this as a sole (no pun intended) crop. On a large scale, Egyptian farmers grow fish in collection ponds in the Nile Delta, as displayed in this astronaut photo:

Astronaut photo from the International Space Station of dikes in water brought in from the Mediterranean Sea; water has a grayish color in this image; the diffuse black patches in the image center are cloud shadows.

This part of the Delta contains about a half of Egypt's aquacultural industry. That industry sprang up after the Aswan Dam was built. The dam's ecological impact has diminished the nutrients that helped to feed fish in the Nile River. The managed Delta fisheries are the replacement of the depleted river supply.

Source: http://rst.gsfc.nasa.gov