Seeing Forests Through the Trees
By Gabrielle J. Corcoran
and A. Mark Dalessandro
James B. McGraw, the Eberly Family Professor
of Biology at WVU, was conducting research on plant species in
the Appalachian Mountains when he decided there had to be a better
way. McGraw, interested in tracking individual trees over time
in order to predict the future of forests, wanted to increase
the number of trees he could examine.
In the past, a plant biologist like McGraw or a forester would
need to walk through a forest, using tape measures and other
low-tech tools to collect information. The scientist collected
samples from a limited area, then drew conclusions about the
entire forest based on the sample.
Although still very useful, these methods allow only for examination
of trees by the hundreds. McGraw wanted to examine millions.
"In our profession, population biology, we tend to work
with a few hundred trees or a thousand trees because that is
all we can do from the ground," McGraw says. "The real
question is: What are a million trees doing? You can't extrapolate
to a very large area from 500 trees."
Remote sensing, an often highly technical scientific approach
to collecting and interpreting specific information without actual
contact with the object being examined, has become a widely used
method for obtaining information about the earth's surface. Images
of the earth are taken from satellites or by using other technologies.
Then, a computer processes the data and isolates the desired
information.
Using remote sensing to examine trees enables the researcher
to assess the resources of an entire forest and collect information
about varieties, sizes, and total numbers of species in the forest.
Researchers in many different fields and industries use remote
sensing of data to obtain information they were not able to acquire
in the past. The Robert Mondavi Winery in California's Napa Valley,
for example, is collaborating with NASA to use digital remote
sensing to help winemakers harvest their grapes at the peak of
ripeness.
McGraw thought he could use a similar approach to increase his
sample from hundreds of trees to millions, and as a result more
accurately draw conclusions about environmental change.
"Applying remote sensing to my research would be particularly
helpful when you have environmental problems like global climate
change and you expect the changes to be pretty subtle. If you
predict only a one percent change in mortality, detecting that
on 500 trees is very difficult because you only expect five trees
to die," explains McGraw. "But if you look at five
million trees, it becomes much easier to detect."
Applications of this technology include the detection of rare
or endangered species, which can lead to their conservation and
restoration. Identifying the presence of pests before a fatal
attack is another valuable function.
Computer programs capable of analyzing an image of a forest and
singling out particular characteristics about a species in that
image have been under development for more than 20 years. But
these programs are designed for northern forests where the species
are primarily coniferous and contain fewer varieties. Forests
in the southern Appalachians, such as those under observation
by McGraw, contain many more varieties of coniferous and deciduous
tree species, making differentiation between individual species
much more difficult.
This fact did not deter McGraw. If a program capable of analyzing
an image of more complex forests did not exist, he and his fellow
scientists would create one.
Unaware of the complexity of developing such a program, McGraw
approached Timothy Warner, a WVU associate professor of geology
and geography, for help. Warner, a native of Zimbabwe, is an
expert in the spatial analysis of remotely sensed data. He told
McGraw the project was not viable.
He was wrong.
"I had only worked on much coarser scales where you can't
differentiate between trees, and therefore I was biased,"
says Warner. "McGraw was limited in his knowledge of remote
sensing, but not in his thinking."
Despite his doubts, Warner did not completely close his mind
to the challenge. He assigned Thomas Key, a geography graduate
student, to investigate the project further. Key's research indicated
that there were ways to identify individual species. So, Warner
and McGraw decided to seek funding to develop the required technology.
NSF-EPSCoR (National Science Foundation Experimental Program
to Stimulate Competitive Research), a program to enhance federal
funding to states in need, approved their proposal and provided
them with $50,000 to begin work on the project. Because the project
was viewed as risky, they would need to make substantial progress
before seeking further funding.
Now, after five years of research, including a three-year, $332,000
grant from the NSF, the WVU project has made breakthrough advances
in remote sensing.
"We are taking a new perspective on plant population biology,"
McGraw says. "Instead of walking through the forest, we
are looking at a complex forest from above and collecting data
that was previously only possible from the ground."
Writing a computer program capable of identifying individual
species located in forests as diverse as those of the southern
Appalachians might have been impossible without the help of Tomas
Brandtberg, who earned his doctorate at the Swedish University
for Agricultural Sciences in Uppsala.
Brandtberg had just completed his dissertation on high-spatial
resolution remote sensing and was looking for work. The timing
could not have been better for Warner and McGraw, who were searching
for someone with the expertise Brandtberg displayed in his dissertation,
which received an award from the Royal Society of Sciences in
Uppsala. In addition, Brandtberg's extensive knowledge of forestry
enabled him to appreciate the difficulties he would face in developing
software to analyze a complex forest.
Warner hired Brandtberg as a post-doctoral research associate
and research assistant professor of geography. "We are extremely
fortunate to have him here because he is on the cutting-edge
of high resolution remote sensing research," McGraw says.
"His work is some of the best in the world."
With Brandtberg hard at work writing computer algorithms, Warner
and McGraw have developed methods to identify individual species
once the computer program makes the distinction between the different
plants. One method, based on tree phrenology—the timing
of leafing out and the coloration of tree foliage—can be
used to recognize specific species during a particular time of
year.
"Dr. Brandtberg is making considerable progress delineating
the edges of individual trees and then using these color patterns
to distinguish species," McGraw says.
Studying the leafing-out of trees has proven to be effective,
but it assesses only the forest's top layer. So, a LIDAR (light
detection and ranging) instrument that detects plants beneath
the forest's canopy has been incorporated in the investigation.
McGraw explains that, by using signals that travel through the
leaves, LIDAR can detect both the location and amount of vegetation.
"The LIDAR method is very significant because it will also
construct the three-dimensional shape and height of the tree,"
Warner explains. "It can do this by measuring the tree at
several points in order to reconstruct its actual shape and potentially
reveal what lies beneath."
Aerotech, a remote sensing company that
specializes in LIDAR and digital image surveys, provided the
data used for this research free of charge through a procedure
that typically costs about $50,000. Aerotech, which was experiencing
similar struggles in devising methods to identify particular
trees from LIDAR data, sees potential profits in a commercial
spin-off that could come from its collaboration with the WVU
researchers.
Forestry companies, for example, may be able to use the technology
to inventory their resources and plan selective cutting to maximize
long-term forest health. McGraw explains that land managers will
be able to classify forest communities in greater detail than
previously possible, know where rare species are found, assess
changes in forest composition, and use the detailed canopy maps
to guide management plans.
"The hope is that we both benefit," Warner says. "For
us, access to data from a state-of-the-art system, acquired by
professionals who have tremendous expertise in capturing this
data, is invaluable. For Aerotec, there is the possibility that
our work will improve an aspect of their work and open new markets
in forest mapping."
McGraw and Warner are excited about the potential applications
of their work, including the possible restoration of the American
chestnut tree. This species, once one of the most valuable of
the eastern deciduous trees, was eradicated by the chestnut blight
in the early part of the 20th century. However, a few adult trees
may have survived, and could potentially be located through remote
sensing.
Hickory trees, another valuable species found in the Appalachian
forests, are also declining in numbers, but only in particular
regions. By examining environmental factors of the different
regions, researchers could possibly identify which factors are
causing the decline of this species.
Rick Landenberger, a post-doctoral research associate and a research
assistant professor of biology, has been working on the tree
delineation project and another project McGraw and Warner began
last fall involving censusing natural populations of silversword
in Hawaii's Haleakala Crater. Remote sensing, in conjunction
with on-the-ground monitoring of this unique, long-lived plant
species, may allow for a more accurate and complete census of
the federally protected silverswords. Landenberger has coordinated
the data acquisition and processing on the silversword project,
which McGraw plans to continue working on during a sabbatical
this fall.
M. Duane Nellis, dean of WVU's Eberly College of Arts and Sciences
and a professor of geography, has also been collaborating with
Warner and McGraw on vegetation inventorying and monitoring projects.
"We are especially pleased to be working with Duane,"
Warner says. "He has an incredible reputation in the discipline
of remote sensing, and brings a great deal of knowledge and expertise
to the project."
Although Warner and McGraw have begun to employ their research
in a number of projects, their work is nowhere near complete.
They are uncertain if each species has a unique color signature,
which will be necessary for the leafing-out method to be effective.
They have begun examining the possibilities of using other attributes,
such as branching patterns, to identify species. Continuing to
refine the technologies used to locate individual trees in the
images and identifying them by species is their primary research
focus for the next several years.
"These problems are not completely solved at this point,"
says McGraw. "Another area in development is taking images
in different years and determining whether we can pick out the
death and birth of canopy trees. This kind of information could
lead to predictive models of forest change."
"There are all these different aspects that need to be examined
here," Warner says. "There is the issue of the rare
and endangered species, the examination of subtle changes in
population of common species, the LIDAR and structuring the three-dimensional
shapes of the trees, as well as the color patterning. Even though
we have a very focused number of questions that we are working
on at this moment, it really is a huge research agenda."