22
phy and found that variation in crown size was primarily related to the range of the variogram,
whereas crown closure affected the sill. Cohen and
Spies (1992) showed high correlation between canopy tree density and texture in SPOT HRV lO-m
panchromatic imagery.
Low-resolution optical geometrical models have
been used with some success to estimate tree and
shrub density over large areas. For example, Wu
and Strahler (1994) inverted an optical geometrical
forest canopy reflectance model to estimate crown
size and density in various conifer stand types in
the northwestern United States. The inversion procedure systematically underestimated both crown
size and density, but the estimates were highly correlated with field measurements (r > 0.9 for
density).
Cover and Leaf Area
Direct Methods
In general, cover refers to the vertical projection of
vegetation onto the ground when viewed from
above. (Unfortunately, crown density is often used
synonymously with cover in the forest mensuration
literature [Philipson 1997]). Point and line intercepts are the most common means for directly measuring cover, and there are many variations on the
general approach (reviewed by Bonham 1989). Because cover is so tedious to measure directly, visual
cover estimation has been widely used for survey
and phytosociological studies (Mueller-Dombios
and Ellenberg 1974), although not without bias
(Gotfryd and Hansell 1985). McAuliffe (1990) devised a method for rapid and less-biased estimation
of cover in sparse shrublands based on using logarithmic series of canopy density and diameter
classes.
Indirect Methods
On the ground, cover can be measured indirectly
based on the point intercept method using optical
or electromagnetic devices, such as an upward-or
downward-pointed "moosehorn" with bubble levels (Mueller-Dombios and Ellenberg 1974) or an
upward-looking gridded concave mirror (spherical
densiometer, Lemmon 1956). Alternatively, gap
fraction methods can be used to estimate cover in
both low and tall vegetation.
Frank W. Davis and Dar Roberts
Many techniques have been developed to estimate canopy cover from aerial photographs. Most
are based on either subjective visual estimation, a
sampling approach (e.g., dot grids, line intercept),
or area estimation (e.g., planimetering of tree
crowns) (Husch et al. 1982). Aerial photographs
may also be used as one stage in a multistage estimation procedure based on the measured relationship between cover measured in ground samples
and estimates of the same areas made from largescale aerial photographs (e.g., Pitt et al. 1996). Image processing methods have been used with mixed
success to automatically delineate tree or shrub
crowns in H-resolution digital imagery in order to
measure crown size, cover, or density (Gougeon
1995).
Multispectral aircraft and satellite imagery has
been extensively applied to estimate canopy cover.
At scales ranging from 5 to 60 m, vegetation structure is expressed as subpixel mixtures of canopy
components, including shadows, green leaves,
branches, trunks, and exposed soilllitter, as well as
interpixel variation in crown illumination due to
stand scale heterogeneity in crown shape, density,
and tree height. Analysis techniques at subpixel
scales can be placed into two broad categories,
ratio-based indices and linear techniques. Ratiobased indices primarily include the standard ratio
(near infrared/red) and the normalized difference
vegetation index (NDVI) (Tucker, 1979; Sellers
1985). They have been used to quantify LAI,
FPAR, and percentage cover and have been highly
successful in agricultural applications and in some
analyses of forest structure (see Chapter 3). Although most applications have used data from operational orbiting LANDSAT or NOAA sensors,
forest LAI has also been measured with high accuracy from aircraft-borne imaging spectrometers
using simple regression models and spectral indices
(Gong et al. 1995). Time series of NDVI images
generated from global AVHRR data sets have been
applied to monitor seasonal and interannual
changes in canopy greenness and ecosystem primary production (Reed et al. 1994, reviewed by
Defries et al. 1995). Recently, Myneni et al. (1997)
demonstrated a relatively simple and robust algorithm for retrieving LAI and FPAR at continental
to global scale.
Linear techniques include indices, such as the
perpendicular vegetation index (PVI); (Richardson
phy and found that variation in crown size was primarily related to the range of the variogram,
whereas crown closure affected the sill. Cohen and
Spies (1992) showed high correlation between canopy tree density and texture in SPOT HRV lO-m
panchromatic imagery.
Low-resolution optical geometrical models have
been used with some success to estimate tree and
shrub density over large areas. For example, Wu
and Strahler (1994) inverted an optical geometrical
forest canopy reflectance model to estimate crown
size and density in various conifer stand types in
the northwestern United States. The inversion procedure systematically underestimated both crown
size and density, but the estimates were highly correlated with field measurements (r > 0.9 for
density).
Cover and Leaf Area
Direct Methods
In general, cover refers to the vertical projection of
vegetation onto the ground when viewed from
above. (Unfortunately, crown density is often used
synonymously with cover in the forest mensuration
literature [Philipson 1997]). Point and line intercepts are the most common means for directly measuring cover, and there are many variations on the
general approach (reviewed by Bonham 1989). Because cover is so tedious to measure directly, visual
cover estimation has been widely used for survey
and phytosociological studies (Mueller-Dombios
and Ellenberg 1974), although not without bias
(Gotfryd and Hansell 1985). McAuliffe (1990) devised a method for rapid and less-biased estimation
of cover in sparse shrublands based on using logarithmic series of canopy density and diameter
classes.
Indirect Methods
On the ground, cover can be measured indirectly
based on the point intercept method using optical
or electromagnetic devices, such as an upward-or
downward-pointed "moosehorn" with bubble levels (Mueller-Dombios and Ellenberg 1974) or an
upward-looking gridded concave mirror (spherical
densiometer, Lemmon 1956). Alternatively, gap
fraction methods can be used to estimate cover in
both low and tall vegetation.
Frank W. Davis and Dar Roberts
Many techniques have been developed to estimate canopy cover from aerial photographs. Most
are based on either subjective visual estimation, a
sampling approach (e.g., dot grids, line intercept),
or area estimation (e.g., planimetering of tree
crowns) (Husch et al. 1982). Aerial photographs
may also be used as one stage in a multistage estimation procedure based on the measured relationship between cover measured in ground samples
and estimates of the same areas made from largescale aerial photographs (e.g., Pitt et al. 1996). Image processing methods have been used with mixed
success to automatically delineate tree or shrub
crowns in H-resolution digital imagery in order to
measure crown size, cover, or density (Gougeon
1995).
Multispectral aircraft and satellite imagery has
been extensively applied to estimate canopy cover.
At scales ranging from 5 to 60 m, vegetation structure is expressed as subpixel mixtures of canopy
components, including shadows, green leaves,
branches, trunks, and exposed soilllitter, as well as
interpixel variation in crown illumination due to
stand scale heterogeneity in crown shape, density,
and tree height. Analysis techniques at subpixel
scales can be placed into two broad categories,
ratio-based indices and linear techniques. Ratiobased indices primarily include the standard ratio
(near infrared/red) and the normalized difference
vegetation index (NDVI) (Tucker, 1979; Sellers
1985). They have been used to quantify LAI,
FPAR, and percentage cover and have been highly
successful in agricultural applications and in some
analyses of forest structure (see Chapter 3). Although most applications have used data from operational orbiting LANDSAT or NOAA sensors,
forest LAI has also been measured with high accuracy from aircraft-borne imaging spectrometers
using simple regression models and spectral indices
(Gong et al. 1995). Time series of NDVI images
generated from global AVHRR data sets have been
applied to monitor seasonal and interannual
changes in canopy greenness and ecosystem primary production (Reed et al. 1994, reviewed by
Defries et al. 1995). Recently, Myneni et al. (1997)
demonstrated a relatively simple and robust algorithm for retrieving LAI and FPAR at continental
to global scale.
Linear techniques include indices, such as the
perpendicular vegetation index (PVI); (Richardson