By Jim Geiger, Center for Urban Forest Research,
Pacific Southwest Research Station, USDA Forest Service (www.fs.fed.us/psw/programs/cufr),
Davis, Calif.
Imagine flying over a community forest and taking
a picture that allows you to identify and map species. Then imagine the
possibilities and cost savings if this method could eliminate the need
for field surveys.
Establishing accurate tree type and species parameters is critical for
urban forest management, cost/benefit analysis and urban planning.
Traditionally, urban forest managers have obtained these parameters by
analyzing field surveys. However, a recent study at the Center for Urban
Forest Research with NASA’s Airborne Visible InfraRed Imaging
Spectrometer (AVIRIS), under the leadership of University of
California-Davis’s Dr. Qingfu Xiao, suggests there may be a much cheaper
and equally effective alternative.
Understanding the Urban Forest
To understand how urban forests function and to estimate the value of
their environmental services, it’s important to identify properties
related to urban forest structure and composition. Also, understanding a
forest’s structure provides other information useful to urban managers,
such as planning tree pruning and removal, as well as insect or disease
control activities.
Basic information required to describe urban forest structure includes
tree numbers, spatial distributions, species
composition, dimensions and growing conditions.
Traditional field surveys are expensive and time consuming, and require
periodic updates to remain valid. Aerial photograph interpretation has
been used successfully, but is slow and expensive to conduct the
mapping on a large scale.
What About Infrared Imagery?
Vegetation has unique spectral reflectance characteristics, which
makes infrared imagery so attractive. Vegetation has a high
absorption rate in red wavelengths and a strong reflectance rate in
near-infrared wavelengths. This allows researchers to separate
vegetation from other ground-surface covers because nonplant
material absorbs and reflects infrared at a different rate.
Differences in foliage, branches and architecture among tree species
provide information that allows AVIRIS to uniquely identify them.
Differences in canopy architecture, such as leaf area density, leaf
and branch angles, leaf shape, internal anatomy, and leaf and branch
surface roughness, cause individual tree species to reflect
differently.
Additional Methods
The Normalized Difference Vegetation Index, red-edge and other band
ratio methods also can be used to separate vegetation types.
However, these simple methods can’t be used to identify tree species
because they don’t capture the unique spectral characteristics of
each species. Another method, texture analysis, works well in
natural forest mapping to identify species, but it doesn’t work well
in the urban forest because urban tree species are too similar in
texture.
Satellite-based imagery—Landsat Thematic Mapper (TM) seven-band,
30-meter data; SPOT four-band, 20-meter data; and multispectral
imagery from commercial satellites such as IKONOS and QuickBird—have
significantly improved the accuracy of identifying vegetation,
especially estimates of dominant tree species. However, the accuracy
in urban settings becomes a problem because urban areas are a mosaic
of many different species, land uses and man-made structures, each
of which has different spectral reflectance characteristics.
Unlike trees in rural forests, which tend to
form continuous canopies, trees in urban settings are often single
trees or isolated groups. Background influence, such as soil and
shadow, makes the problem of characterizing trees by remote sensing
even more difficult. In such cases, high spatial resolution of
remotely sensed data is important for mapping individual trees.
Why AVIRIS?
AVIRIS compensates for the variety of backgrounds in urban areas by
delivering calibrated images in 224 contiguous spectral channels,
with wavelengths ranging between 400nm and 2,500 nm. This enriched
spatial and spectral data reduces the resolution problems associated
with broad-band low-spatial resolution sensors, such as Landsat with
just seven channels, and SPOT and the commercial satellites with
just four, thus giving AVIRIS the ability to “see” trees 30 to 70
times better than other methods.
Combining AVIRIS with geographic information
system (GIS) software significantly improves the accuracy of the
AVIRIS results. The spatial location ability of GIS is a standard
method for registering images to base maps. This ability to
accurately locate individual trees using GIS, combined with the
AVIRIS analysis, makes it relatively easy to confirm the AVIRIS
results. Plus, it significantly raises the confidence level when
replicating the procedure in other city areas or nearby regions.
Study Objectives
There were three objectives for the Center for Urban Forest
Research’s recent AVIRIS study:
1. Identify urban tree species by physiognomic type based on their
spectral character as detected by the AVIRIS sensor—i.e., discern
whether they are broadleaf deciduous, broadleaf evergreen, or
conifer types.
2. Identify urban trees by species based on their canopy reflectance
characteristics.
3. Map these urban trees.
The results were checked against ground reference data and compared
with tree information in an existing GIS database. At the tree-type
level, mapping was accomplished with 94 percent accuracy. At the
tree species level, the average accuracy was 70 percent, but this
varied with both tree type and species. Of the four evergreen tree
species, the average accuracy was 69 percent. For the 12 deciduous
tree species, the average accuracy was 70 percent. The relatively
low accuracy for several deciduous species was due to small tree
size and overlapping among tree crowns at the 3.5-meter spatial
resolution of the AVIRIS data.
This means that it’s possible to identify individual tree species
with fairly high accuracy using high spatial resolution (3.5 meter)
AVIRIS data. Therefore, the answer to the question posed at the
beginning of the article is “yes,” it is possible to replace field
surveys with AVIRIS. And when combined with GIS, it adds the ability
to validate the final maps.
The potential value of these data for urban forest applications,
besides species identification, includes estimating tree health
(stress and vigor), leaf area, and canopy cover. In addition to tree
characterization, AVIRIS can be used for characterizing land cover.
For example, it’s possible to classify man-made structures, such as
buildings or type of pavement (porous,concrete, asphalt, gravel), by
the materials used.
AVIRIS data acquired in spring or summer
rather than fall might provide better identification of some species
or additional information about tree condition. For example, data
acquired in both summer and winter seasons could be used to easily
identify deciduous and evergreen trees.
The mix of land cover for street trees also plays an important part
in the outcome. Pixels of most street trees in residential areas
will be mixed with road and/or turf grass. Street trees also will be
mixed with bare soil and/or road in median strips and in some
commercial areas. This mixing reduces the number of possible
combinations and is the greatest reason that accuracy increased in
this study compared with earlier results with less sophisticated
techniques. Because most trees will still be within mixed pixels at
this scale (3.5 meters), increasing spatial resolution of the
hyperspectral dataset could improve the accuracy of tree
identification.
Caveats
This urban forest tree species mapping method has the potential to
improve the accuracy of urban tree mapping while reducing costs
compared with field sampling or other traditional methods. However,
the center also found that it isn’t fully transferable from one city
to another without calibration from ground truthing. Also, using
this method to identify trees in locations other than along the
street may not yield the same results due to the potential for more
complex mixing combinations off the street.
