Forest Applications of Remote Sensing - Lecture Note - Completely Remote Sensing tutorial, GPS, and GIS -
Forest Applications of Remote Sensing

Let's switch attention now to the world's forests, natural and planted and maintained for future harvesting. The distribution and condition of these forests, along with other principal types of vegetation in the global biosphere, can be determined and refined periodically, by analyzing Vegetation Index data acquired by satellites. As an example, here is a plot of the distribution of general forest classes, as they were identified in AVHRR data from NOAA 7 and 9 metsats. Colors that are not identified in the legend include red = shallow, coastal or inland waters; blue = ocean waters; purple = deep ocean; black in the arctic and antarctic regions = not classified.

Generalized map of forest classes plotted worldwide.

One obvious application lies in determining the geographic and areal distribution of major types of sylvan vegetation and their ecological settings. Generally, Landsat and other remote sensing systems can readily distinguish between deciduous and coniferous forests, and can recognize shrublands, savannahs, and various desert ecosystems. But, identifying most tree types at the species level is much more difficult, unless high resolution images are available and the trees have distinctive crowns and leaf shapes that produce fairly explicit spectral signatures. Easier tasks, that have widespread applications, involve discerning defoliation - either in the extreme with tree removal by clearcutting or through denuding trees of their leaves by insects, and assessing forest fire or storm damage.

We all know from experience that vegetation cover varies seasonally. This reaches maximum differences in active growth in the higher latitudes (and altitudes) and is less changed in the equatorial and tropical regions of Earth. Here are two global vegetation cover maps made from MODIS data, with the growth data collected over time frames of weeks:

MODIS maps of global vegetation cover (calculated as the Enhanced Vegetation Index) for two different seasonal periods.

This map indicates that parts of the world have extensive, in places continuous, forest cover. One such region is the Congo in central Africa (where the notion of "jungle" was fostered in the 1930s by the Tarzan movies). This next image shows a nearly complete forest canopy so dense that, in the false color rendition, the tone is a uniform red. That type of cover marks true jungle, matched elsewhere mainly in the Amazon basin described later on this page.

Landsat image of the thick canopy of trees in the jungle of the Congo in Africa; the Congo River crosses the scene.

Look carefully at this image - the effects of the underlying topography is expressed in the canopy by variations in the shading..

Two maps published in 2010 show tree canopy heights worldwide and in more detail for the U.S. MODIS and a Lidarlike instrument were used. The maps do not display areas of small forest patches or groves of trees, but only the continuously forested areas. In the U.S. this coverage is associated mainly with mountainous regions where slopes are too steep for building.

Tree canopy distribution worldwide.
Tree canopy distribution in the U.S.

Tree and Forest Classifications

By now, the question should have crossed your mind as to how well individual tree species (and the individual trees themselves) can be identified by various remote sensors and data processing techniques. Generally, if complete spectra over Vis, NIR, and SWIR regions of the EM spectrum are available, the discriminatiion of species can be accomplished with acceptable accuracy. This set of spectral signatures support that claim:

Spectral signatures for various deciduous and coniferous tree species.

Note that the reflectances for the several coniferous tree species (fir, pine, cedar) are lower than those for deciduous trees (maple, oak, aspen)

For Landsat, the answer to species identification is generally "not as well as often desired" because of both spatial and spectral resolution. In general, at lower resolutions dark evergreens can be separated from deciduous trees in temperate and boreal ecosystems. The pine forests of eastern New Jersey, which we saw earlier in the Introductory Section, are distinguishable from the deciduous forests in that state by their darker reddish-brown tones. We reproduce this image here; note other areas in the scene, such as in the Appalachian segment, that are darker, and thus indicative of the occurrence of coniferous trees:

Landsat image of New Jersey.

This next Landsat7+ ETM study of evergreens and deciduous trees in an area in Estonia shows the former appear darker in both natural and false color images relative to the tone brightness of a stand of deciduous trees; the terms "mire" and "fen" refer to wetlands and bogs. The band data were used to produce a classification - at the type rather than species level - of this scene:

Quasi-natural color image of a scene in Estonia.
False color version of the above.
Cluster classification of the scene.

A similar result is also the case for trees in the Pike National Forest in central Colorado near Hayman. An unsupervised classification of a forest area in this Colorado preserve imaged by the Landsat TM yielded the result shown in the top image below. When ground truth was added to the process, evergreens are shown in blue; aspens in green, clear areas in red. A new map (bottom) in which just the aspens are located, as shown in green, proves that some individual tree speciation can be reliably determined.

Unsupervised classification of part of the Pike National forest in Colorado. A single class, in green, is singled out in this classification; all other classes are rendered black.

Broad categories of vegetation and other land cover types can be classified as indicated in this illustration which shows a fair number of categories - most vegetation-related - in the region around St. Petersburg in Russia.

Land cover map (mostly vegetation) of the region around St. Petersburg, Russia; source does not indicate what sensor was used.

Staying in Europe for the moment, examine next this supervised class map of vegetation in hilly terrain in Austria:

Supervised classification of land cover (7 classes of vegetation) in Austrian hills.

Again, as was true with crops, high resolution imagery greatly improves the ability to identify and locate individual tree species. In this IKONOS 4 meter subscene of a forest near Charleston, South Carolina, individual Tupelo and Cypress trees, and a class of the two mixed, have been pinpointed by classification.

IKONOS image with Tupelo and Cypress tree distributions shown by colors indicated.

Perhaps the best example on this page of the efficacy of remote sensing to classify vegetation and other ground cover types is this series of images showing a forest in Canada that contains conifers (mostly Jackpines), deciduous trees, and fens (bogs). On the left is a map made from ground and aerial photos. Next to it is a surprisingly good Landsat TM classification. The right two images are made from AVIRIS data; the left one is based on using the Leaf Area Index values and the right uses the NDVI approach to calculate the classes.

Four maps of a forested area in Canada; see above.

The above AVIRIS images imply that high resolution hyperspectral data, coupled with adequate ground truth, can be quite effective in classifying trees down to the species level. This study of urban tree classification along the streets of Modesto, California supports that conclusion:

Classification of trees in a Modesto, CA neighborhood; B is a field map made onsite; C is the results of classifying AVIRIS data.

Radar has been used to determine tree types. as shown in this SIR-C investigation of a NASA Supersite at Raco, Michigan on the Upper Peninsula next to Lake Superior. The host Shuttle was STS-59 which flew in April, 1994. The target area was heavily forested, as seen in the multiband color composite:

Color composite made from three radar bands on SIR-C, showing a forested area near Raco, MI.

The blue area is mostly cleared fields. Light greenish-yellow marks red pines; brownish-yellow denotes jack pines, and purple locates deciduous trees.

Using ground truth and other information sources, the above image was classified as shown in the left illustration below. Estimates of biomass, using field data and a mathematical model, are shown in the right illustration.

Species classes and biomass determined from the Raco, MI image.

Both Visible-NIR and radar images have met with moderate success so far in estimating biomass in forests, grasslands, and crops. The degree of success is strongly influenced by the quality of ground truth and the specific model used. But as higher resolution imagery and better models are developed, these results should improve significantly.

An important parameter in determining biomass is the "crown height", i.e., the height to the top of a forest canopy from the ground. ICESat's GLAS laser ( see page 14-14) can make such a determination along a profile line as it orbits over non-polar regions. Here is an example of measured heights in an area in the Rocky Mountains:

Canopy heights in a Colorado forest.

The Amazon Rain Forest; Deforestation

Drained and nourished by the mighty Amazon river and its tributaries, the Amazon rain forest is the largest such tropical jungle in the world. It hosts myriads of plant and animal species unique to the region. Some plants have proved invaluable as sources of chemicals that can be converted into medicines. Until recent decades it was inhabited almost exclusively by native aboriginal peoples. This photo shows a typical stretch of unspoiled jungle canopy:

Forest canopy in the Amazon jungle near Manaus, Brazil.

Clearcutting in parts of the world has become a major problem and is the subject of much environmental concern, because of habitat destruction and biochemical threats to the atmosphere. Clearcutting disturbs the balance between the oxygen and carbon dioxide produced and used in photosynthesis. The systematic attack on rain forests in the tropics, to provide open land for farming and other development, is currently the greatest source of alarm. The top concern is for the vast jungles of the Brazilian Amazon, which transition into savannahs at their edges. Over the years, farmers, ranchers, and other entrepreneurs have cleared about 200,000 square miles (about 10%) of the 1.93 million square miles in the Amazon Basin. Fortunately because of an outcry from the rest of the world, the rate of removal is slowing somewhat.

This image map made by the ASAR (radar) on Envisat shows the Amazon drainage basin to be generally low in elevation and flat. The absence of blocking mountains to the east, so that Atlantic marine moisture can move westward unimpeded, is a key factor in the reason for the Amazon basin forests covering most of the landscape. That, plus the tropical conditions near the Equator, which means warm, steady tempertures (no significant cooling since seasonal changes are minimum) and abundant rainfall, have led over the eons to the build-up of thick, luxurious, species-varied, and persistent vegetation of jungle-like character.

Elevation map made by radar showing the vast lowlands of the Amazon, with higher terrain to the southeast and the lofty Andes to the west.

Radar data from the NASA Scatterometer (NSCAT) (see page 14-12) operating over South America shows (below) the distribution of several general, land-cover classes in the Amazon. In this image-based map, rain forest appears in blue/purple, woodlands and savannah in green or yellow, and farmlands or undeveloped mountain lands in black.

 Radar scatterometer image of northern South America; the Brazilian rainforest stands out from other surface features and is displayed in purple-red.

Where the Amazon forest has not been clearcut, it is extremely dense - a true jungle. This Landsat-1 false color demonstrates the uniformity of the forest canopy.

Landsat-1 image of the Amazon jungle.

This next Landsat TM image is a natural color composite that further illustrates the uniformly dense tropical forests. These forests are the largest continuous tract of thick woodlands anywhere on Earth - and provide a haven for diverse plant and animal species, now threatened with destruction.

Natural color composite of Landsat TM data, showing the uniformly dense and thick tropical vegetation in part of the Brazilian Amazon Basin.

One of the surprises resulting from analysis of multitemporal space-acquired images of the Amazon Basin is that, according to determination of EVI (Enhanced Vegetation Index) (Dry - Wet NearIR images), the Amazon forests are "greener" in the Dry Season than in the rainfall-deluged Wet Season:

EVI map of the Amazon Basin.

The most plausible explanation for this: During the Dry Season, when increased sunlight means greater photosynthesis, the trees receive adequate moisture from roots within the saturated soils, thus causing more leafy growth.

The extensive, dense woodlands of the Amazon have made them targets for systematic cutting down, either to supply wood for building or to clear land to convert it into farms or other use. This image below shows the early stage of deforestation in the Amazon:

Initialization of forest removal in the Amazon.

One of the most active programs for converting rain forests to cultivated land persists within the State of Rondonia in the southwestern Amazon. In fact, this area is justly cited as the prime example of land stripping. The next three images support this conclusion. The first (left) image is a Landsat TM subscene taken in 1986. The neat rows or strips in blue are cleared land amidst the surviving forest. In the right image is a Synthetic Aperture Radar (SAR) image taken by the Japanese JERS-1, in which areas of clearcutting show in red.

False color Landsat TM image of the rain forests in the State of Rondonia, South America.
SAR image, taken by JERS-1, of the rain forests in the State of Rondonia, South America.

This pattern is among several shapes present in this image of Bolivian deforestation:

Another view of deforestation patterns in eastern Bolivia.

The rectangular forest stripping pattern is common. Another pattern, which could be called stellate, involves deforestation outward from a common center, as shown in the astronaut photo of an area in eastern Bolivia, and then in a SPOT image:

Stellate deforestation pattern; eastern Bolivia, photographed from the Space Shuttle.
SPOT image of stellate deforestation.

Radar imagery, which can penetrate cloud cover (see page 8-5), is especially effective for short-term monitoring of changes in deforestation. This next image uses two JERS-1 SAR images taken 15 months apart to pinpoint the enlarged edges of cleared strips. Preserved rain forest is shown in white. Deforestation as of July 1992 appears in yellow, and additional stripping up to October 1993 is marked in red. While the changes are not gross, the encroachment on remaining natural forests is obvious

Composite JERS-1 SAR image of a small part of the clearcut rainforest in Rondonia, Brazil; cutting into 1992 in yellow and removal from then into 1993 in red.

Rondonia has become a sort of "type locality" for typifying and monitoring large-scale deforestation. This GOES image shows just how large the clear-cut area is, as the light-toned deforestation is set out against the outline of the State of Florida to elucidate the scale:

GOES satellite image of part of Rondonia, in Brazil, with the outline of Florida superimposed to illustrate the extent of deforestation (white).

Removal of this much of the rain forest produces an area capable of locally modifying the weather, and overall climate. On the left is a TRMM image indicating a rise in temperature; on the right the TRMM image shows that rainfall has actually increased because the extra heat is causing more evaporation above the scarred landscape:

Temperature (in Kelvins) map averaged for a short period of time in the Rondonia clearcut area appearing above. Rainfall (in centimeters) over the same period.

Over the last 30 years, imaging of the rain forests of South America from various satellites has accumulated a lot of history about deforestation as it progresses. This next image uses color coding to indicate removal and some reforestation in the Tierras Bajas jungle in the Santa Cruz region of eastern Bolivia:

Clearing of forest areas in the Santa Cruz region of Bolivia, using observations from a variety of different satellites along with aerial photography.

Many environmentalists, especially the "greens", have deplored what appears to be wanton destruction of the South American rain forest. Although the percentage of cleared land remains small (<10%), strong efforts continue to attempt to curb these losses. Yet there are certain mitigating circumstances that should be considered in optiing for blanket restrictions. At the heart of the counterarguments are these facts: 1) population growth will inevitably lead to incursions into the rain forests as food stuffs and other materials are needed; 2) the nature of the soil (e.g., widespread lateritic soils that are poor in nutrients) over much of the region is poor for long-term use; 3) farms established on such soils may be usefully productive for only a few years; 4) when production becomes limited and unprofitable, the custom is to burn off the last crop remnants, abandon the land, and strip or burn off forests to obtain new land for similar short-term use. The bottom line: reworking parts of the rain forests is inevitable; what is needed is a master plan and regulations to restore natural forest growth in the abandoned lands. Conservation laws should be written to restrict deforested areas to at most no more than 20% of the rain forest, by maintaining land use equilibrium through procedures that put back natural conditions through mandatory reforestation (and also including attempts to improve soil quality by appropriate fertilization).

Other Examples of Clearcutting

At a smaller scale, but nevertheless subject to continuing debate, is the stripping of whole tracts of woodlands in temperate and boreal forests to provide lumber and paper pulp. Loss of the redwoods in the western U.S. coastal forests is a typical situation. While current laws require reforestation of most of these lands, the rate of regrowth in some regions is not keeping pace with removal. Some of these forestry applications are nicely portrayed with several Landsat subscenes (below) of heavily timbered areas of Canada, as prepared by the Canadian Centre for Remote Sensing. The first two cover an area near Port Renfro on Vancouver Island in British Columbia, in part of the Pacific Coastal Ranges with fir, spruce, and some deciduous trees. In this next subscene are a pair of TM images, each 7 x 15 km (4.3 x 9.3 miles), taken on July 17, 1984 (left) and June 19, 1991.

Port Renfro, British Columbia shown in these Landsat TM subscenes on July 17, 1984 (left) and June 19, 1991; areas clearcut during this interval shown in pinkish-red.

Each is processed with TM Band 3 = blue, Band 4 = green, and Band 5 = red. This color scheme makes active vegetation green. Clearcutting of the mixed forest is in pinkish-red. Mature vegetation is darker green and regenerated forest appears in light green.

Another way to carry out this change detection is to use two bands from one year and one band from the other year in making the color composite. Note this combination in the following image:

The Port Renfro, B.C. Landsat subscene again but with 1984 images shown in red and green and the 1991 image rendered in blue.

In this 15 x 15 km (9.3 x 9.3 miles) version, TM Bands 4 and 5 from 1984 are assigned to green and red, and Band 5 from 1991 is shown in blue. Once again, mature forest is rendered in dark green and reforested land in light green but here pink denotes areas that were recently deforested in 1984 and blue refers to areas that were clearcut between 1984 and 1991 but have yet to make a strong recovery.

We shift now to part of the boreal forestlands that are widespread on the Canadian Shield. The area (15 x 15 km, 9.3 x 9.3 miles) is in an isolated timber wilderness near Lac Nemiscau in northern Quebec:

False color composite of a 1991 Landsat subscene covering the Lac Nemiscau, northern Quebec boreal forest and tundra; see text for letter explanations.

In this August 20, 1991 subscene, TM Bands 5, 4, and 3 have been assigned to red, green, and blue, respectively. The Riviere de Rupert and associated water offshoots appear in black. Shades of bright green are areas of recent, partially recovered fire burns (B in red lettering) or regenerated forest (at R). Purple-gray denotes coniferous forests (at C) and orange is deciduous forest (at D). Small areas of wetlands and bogs (at W) show as pink; M refers to mixed forest, and H indicates heath (evergreen shrubs, similar to Scottish heather) with trees. Try to pick out the power line running to the north. This false color rendition is almost as graphic and definitive as a supervised classification, because of the incorporation of the longer wavelength, TM Band 5.

Staying in Canada which has a vast area of boreal forest, let's look at another multitemporal example of change as monitored by the multiple bands on SIR-C radar:

A forested area near Prince Albert, in the province of Saskatchewan, Canada; SIR-C image: X band = blue; C band = Green; L band = red.

There are notable changes between April and October of the same year. In April Old Jack Pine is purple but red in October; Spruce and Aspen are red and green in April and blue in October. A small clearcut area made during the summer is also blue in the October image.

The Terra satellite is contributing in many ways to better our knowledge of forests and other vegetation biomes. It, too, provides striking images of deforestation. Perhaps no other part of the U.S. shows the nature and magnitude of clearcutting than the Cascades (and nearby Coast Ranges) in the Pacific Coast states. Here is an ASTER image that shows recently clearcut forests in red (not the actual color of the denuded land, but chosen to emphasize their widespread location). Replanted earlier clearcuts appear in light green while old forest is in darker green; the blue is residual snow surviving into this May, 2000 subscene.

ASTER image of clearcut forest in the Cascade Mountains of Oregon.

This photo shows a ground scene in the Oregon Coast Range foothills in which a once-heavily forested surface has been largely denuded by clearcutting.

Clearcut land in Salem County, Oregon.

With the advent of high resolution space imagery, details of the clearcuting can now be deciphered for small areas. This IKONOS-2 image is of the Copper Mountain area of the Colorado Rockies. With its 4 m resolution, a texture defined by individual evergreens can be made out. Logging roads and cleared sections are readily defined for easy mapping.

IKONOS image of clearcut forest in the Colorado Rocky Mountains.

Nature has its own way to accomplish "clearcutting". For example, hurricanes can down trees or strip off foliage on a grand scale. Hurricane Hugo slammed into South Carolina just north of Charleston in 1989. Millions of trees were severely damaged. By comparing Landsat TM imagery from the year before with an image taken soon after the hurricane hit, areas of major damage could be mapped, as shown in red in this illustration:

Hurricane damage from Hugo.

We insert here an oddity that has something to do with trees. In northern Kazahkstan, local farmers have developed for centuries a land use practice in which each farm is separated from its neighboring farms by a fence along which trees are planted. In this Landsat-7 ETM+ image taken in early spring, the high walls stand out but the trees have yet to leaf:

Farmland in Kazahkstan, with walls between farms that in summer show rows of leafed trees.

There is an esthetic side to watching tree foliage from space. In Fall, New England is famed for its marvelous leaf colors, bringing millions of tourists to witness this panorama of visual beauty. Such imagery helps TV Weather commentators in advising residents in their listening area as to the optimum times and places to view the countryside. This Terra MISR image pair of northern N.E. (western Maine into New Hampshire and eastern Vermont) portray almost true color views of foliage in late August 2000 when the forests were near their height of greeness (left) and past their peak (right) - most leaves are now dropped - in mid-October 2000:

Northern New England, imaged in August and October 2000 by MISR.

Let's also look at a companion MISR image showing southern New England. Here, being further south, the image shows more of a red tone, in keeping with the fact that the leaves (red and yellow in color) have not yet fallen.

MISR near-natural color image of southern New England into New Jersey in October 2000.

With this overview of forest applications, we can proceed through a case study of the use of space imagery to monitor beetle infestation in evergreens (specifically, pines).