Robust Forest Cover Indices
for Multispectral Images
Sarah J. Becker, Craig S.T. Daughtry, and Andrew L. Russ
Abstract
Trees occur in many land cover classes and provide sig-
nificant ecosystem services. Remotely sensed multispectral
images are often used to create thematic maps of land cover,
but accurately identifying trees in mixed land-use scenes is
challenging. We developed two forest cover indices and proto-
cols that reliably identified trees in WorldView-2 multispectral
images. The study site in Maryland included coniferous and
deciduous trees associated with agricultural fields and pas-
tures, residential and commercial buildings, roads, parking
lots, wetlands, and forests. The forest cover indices exploited
the product of either the reflectance in red (630 to 690 nm)
and red edge (705 to 745 nm) bands or the product of reflec-
tance in red and near infrared (770 to 895 nm) bands. For two
classes (trees versus other), overall classification accuracy
was >77 percent for the four images that were acquired in
each season of the year. Additional research is required to
evaluate these indices for other scenes and sensors.
Introduction
Trees occur in forest and non-forest land cover classes and
provide significant ecosystem services, including microclimate
regulation, watershed protection, wildlife habitat, and recre-
ational uses. Trees and their locations within a landscape are
also important for such diverse applications as planning routes
for utilities, roads, and trails, monitoring land cover changes,
and modeling environmental quality and quality of life. Re-
motely sensed images are often used to create thematic maps
of land cover at a range of spatial and temporal scales (Foody,
2002), but accurately identifying trees in mixed land-use
scenes is often challenging because tree cover is easily conflat-
ed with other types of vegetative covers in multispectral data.
The focus of this research is to address this challenge with a
methodology for distinguishing forest cover from other land
covers (including vegetative covers) using multispectral data.
The remainder of this section reviews spectral proper-
ties of vegetation and remote sensing approaches for detect-
ing trees to support understanding how forest cover differs
spectrally from other land covers. The next section details the
materials and methods used to conduct this research, includ-
ing a description of the indices developed to distinguish
forest cover from other land covers and how accuracy was
measured, followed by a section that describes results from
testing the new indices, leading to a discussion and sugges-
tions for future research, and a final section with conclusions.
Spectral Properties of Vegetation
Multiple factors influence the spectral properties of for-
est and other vegetation matter, creating a challenge for
distinguishing between them in satellite imagery. When solar
radiation interacts with matter, it may be reflected, transmit-
ted, or absorbed. The spectral reflectance of vegetation cano-
pies is determined by: (1) spectral properties of the canopy
elements; (2) canopy structure; (3) background reflectance;
(4) illumination and view directions; and (5) atmospheric
transmittance (Bauer, 1985). When vegetation density is high,
leaves are often the primary scattering elements and the back-
ground contributes little to overall canopy reflectance. How-
ever, when vegetation density is low, background reflectance
significantly influences canopy reflectance.
1. The spectral properties of canopy elements (i.e., leaves,
stems, inflorescences) are determined primarily by the
concentrations of chlorophyll and other pigments in the
visible (400 to 700 nm), leaf structure in the near infrared
(700 to 1,200 nm), and amount of water in the shortwave
infrared (1,200 to 2,000 nm) wavelength regions (Knipling,
1970). Physiological and morphological changes occur as
leaves expand, mature, and senesce, which significantly
affect leaf spectral properties (Roberts
et al.
, 1998) and
could aid in distinguishing between trees and other veg-
etation. Nutrient deficiencies, water deficits, and damage
by insects and diseases also affect the spectral properties
of leaves (van Leeuwen, 2009).
2. Although the spectral properties of most healthy green
leaves are roughly similar, canopy structure describes
how individual canopy elements (leaves, stems, etc.) are
positioned throughout the canopy and determines how
radiation is transferred within and from the canopy. Dif-
ferences in canopy structure could aid in distinguishing
between forest and other vegetative covers. The geometri-
cal arrangement of these canopy elements in space varies
with species, age, and environmental conditions, such
as nutrient and water stresses, disease, or insect damage,
and wind. Characterizations of the structure of vegetation
canopies often include leaf area index, fraction vegetation
cover, biomass, and leaf angle distribution (Bunnik, 1978;
Daughtry, 1990).
3. The lower boundary layer or background for vegetation
canopies is typically soil, rock, litter, water, or understory
vegetation, which impact reflectance depending on the
canopy structure and density and can differ between forest
and other vegetative covers. Physical factors quite differ-
ent from those of green vegetation determine the reflec-
tance of these backgrounds and are more likely to influ-
ence reflectance in less dense vegetation canopies. Five
general types of soil reflectance spectra based on organic
matter content, texture, and ferric iron absorption have
been identified (Stoner and Baumgardner, 1981). Rocks
often have unique spectral features associated with their
Sarah J. Becker is with the Geospatial Research Laboratory,
Engineer Research and Development Center, U.S. Army Corps
of Engineers, Alexandria, VA 22315
(
).
Craig S.T. Daughtry and Andrew L. Russ are with the
Hydrology and Remote Sensing Laboratory, USDA
Agricultural Research Service, Beltsville, MD 20705.
Photogrammetric Engineering & Remote Sensing
Vol. 84, No. 8, August 2018, pp. 505–512.
0099-1112/18/505–512
© 2018 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.84.8.505
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
August 2018
505
