1. Stand Structure in Terrestrial Ecosystems
and Wiegand 1977), KT transfonn (Kauth and
Thomas 1976; Crist and Cicone 1984), and a diversity of spectral mixture models (Adams et al.
1993; Roberts et al. 1993). All of these techniques
are based on the assumption that reflected radiance
can be modeled as the sum of reflected radiance
weighted by the aerial proportions of components
within the field of view. In the case of the PVI,
vegetation is modeled as the orthogonal distance in
the NIR to red plane from a soil brightness line.
The KT transfonn extends this analysis to include
up to the six bands of LANDSAT TM, yet develops
the orthogonal indices from a library of soil spectra.
Spectral mixture models derive the mixing lines
(analogous to soil brightness lines and axes from
greenness and brightness) either by deriving pure
spectra from the image (image endmembers) or
from a library of measured field or laboratory spectra (reference endmembers). Vegetation structure is
expressed as variation in either spectral fractions or
one of the other linear measures. For example, a
forested ecosystem might be best modeled as a mixture of green leaves, shadows, and exposed
branches or litter (nonphotosynthetic vegetation,
NPV). Early regeneration of conifers might be distinguished from old-growth forest by a decrease in
green leaves, an increase in canopy shadows (gaps),
and an increase in wood as exposed branches and
snags (Heilman et al., 1996).
Radar profilers and SAR (notably L-band) have
established capability to discriminate forest from
nonforest, but topographic effects pose a real problem. In addition, SAR has demonstrated capacity
to classify landcover types (Pierce et al. 1994).
Thus far, the combined use of radar and optical systems for classification and cover estimation has not
received much attention.
Biomass
Direct Methods
For herbs and small shrubs, biomass may be estimated by harvesting and weighing (Bonham 1989;
Chiariello et al. 1989), although large sample sizes
are needed to account for usually high local spatial
variance. Point intercept methods have also been
applied, although the quality of the estimates varies
depending on plant growth fonn (Frank and
Mcnaughton 1990). For some herbs and shrubs, the
23
notion of reference units has proven useful. In this
approach, small units of the plant are sampled and
weighed, and then the number of reference units in
the larger sample is counted (Andrew et al. 1981;
Carpenter and West 1987).
Nondestructive methods for herbs, shrubs, and
trees depend on allometric relationships between
biomass and more readily obtained direct measures
of cover, LAI, height, basal area, canopy diameter,
or volume (reviews in Causton 1985; Bonham
1989; Husch et al. 1982; Etienne 1989). Thousands
of allometric models have been developed for individual species, structural components, and environments. For example, Means et al. (1994) compiled over 1150 allometric equations for estimating
biomass and other components of structure for
common plant species growing in the western
United States. They also produced public domain
software
(BIOPAK,
http://www.fsl.orst.eduJ
rogues.tsuga/meansj/biopak.htm) to facilitate modeling based on these equations.
Indirect Methods
In herbaceous vegetation, biomass can be highly
correlated with LAI, and thus gap fraction methods
may be used to retrieve biomass based on fitted
allometric equations. In shrublands, crown height,
diameter, or volume may be estimated by ocular
methods or from large-scale aerial photographs,
and then dimensional equations applied to predict
biomass (reviewed by Etienne 1989). Total aboveground forest biomass has been estimated through
a variety of techniques based on ocular estimates
of stand basal area, height, and/or density, and commercial inventory stand and stock tables (Husch et
al. 1982). Stereo aerial photographs have been used
extensively to estimate timber volume based on forest type, stand height, and crown diameter and/or
closure (reviewed by Howard 1991).
Some traditional methods of biomass estimation
exploit a multistage sampling approach to reduce
the variance in biomass estimates obtained over
large areas. These approaches incorporate ocular
methods, aerial photographs, or satellite imagery at
one or more scales to estimate biomass, and then
measure biomass directly by harvesting or other
techniques from a smaller number of sub samples
of the same area (Bonham 1989; Heller and Ulliman 1983). These approaches can be extremely ef-
and Wiegand 1977), KT transfonn (Kauth and
Thomas 1976; Crist and Cicone 1984), and a diversity of spectral mixture models (Adams et al.
1993; Roberts et al. 1993). All of these techniques
are based on the assumption that reflected radiance
can be modeled as the sum of reflected radiance
weighted by the aerial proportions of components
within the field of view. In the case of the PVI,
vegetation is modeled as the orthogonal distance in
the NIR to red plane from a soil brightness line.
The KT transfonn extends this analysis to include
up to the six bands of LANDSAT TM, yet develops
the orthogonal indices from a library of soil spectra.
Spectral mixture models derive the mixing lines
(analogous to soil brightness lines and axes from
greenness and brightness) either by deriving pure
spectra from the image (image endmembers) or
from a library of measured field or laboratory spectra (reference endmembers). Vegetation structure is
expressed as variation in either spectral fractions or
one of the other linear measures. For example, a
forested ecosystem might be best modeled as a mixture of green leaves, shadows, and exposed
branches or litter (nonphotosynthetic vegetation,
NPV). Early regeneration of conifers might be distinguished from old-growth forest by a decrease in
green leaves, an increase in canopy shadows (gaps),
and an increase in wood as exposed branches and
snags (Heilman et al., 1996).
Radar profilers and SAR (notably L-band) have
established capability to discriminate forest from
nonforest, but topographic effects pose a real problem. In addition, SAR has demonstrated capacity
to classify landcover types (Pierce et al. 1994).
Thus far, the combined use of radar and optical systems for classification and cover estimation has not
received much attention.
Biomass
Direct Methods
For herbs and small shrubs, biomass may be estimated by harvesting and weighing (Bonham 1989;
Chiariello et al. 1989), although large sample sizes
are needed to account for usually high local spatial
variance. Point intercept methods have also been
applied, although the quality of the estimates varies
depending on plant growth fonn (Frank and
Mcnaughton 1990). For some herbs and shrubs, the
23
notion of reference units has proven useful. In this
approach, small units of the plant are sampled and
weighed, and then the number of reference units in
the larger sample is counted (Andrew et al. 1981;
Carpenter and West 1987).
Nondestructive methods for herbs, shrubs, and
trees depend on allometric relationships between
biomass and more readily obtained direct measures
of cover, LAI, height, basal area, canopy diameter,
or volume (reviews in Causton 1985; Bonham
1989; Husch et al. 1982; Etienne 1989). Thousands
of allometric models have been developed for individual species, structural components, and environments. For example, Means et al. (1994) compiled over 1150 allometric equations for estimating
biomass and other components of structure for
common plant species growing in the western
United States. They also produced public domain
software
(BIOPAK,
http://www.fsl.orst.eduJ
rogues.tsuga/meansj/biopak.htm) to facilitate modeling based on these equations.
Indirect Methods
In herbaceous vegetation, biomass can be highly
correlated with LAI, and thus gap fraction methods
may be used to retrieve biomass based on fitted
allometric equations. In shrublands, crown height,
diameter, or volume may be estimated by ocular
methods or from large-scale aerial photographs,
and then dimensional equations applied to predict
biomass (reviewed by Etienne 1989). Total aboveground forest biomass has been estimated through
a variety of techniques based on ocular estimates
of stand basal area, height, and/or density, and commercial inventory stand and stock tables (Husch et
al. 1982). Stereo aerial photographs have been used
extensively to estimate timber volume based on forest type, stand height, and crown diameter and/or
closure (reviewed by Howard 1991).
Some traditional methods of biomass estimation
exploit a multistage sampling approach to reduce
the variance in biomass estimates obtained over
large areas. These approaches incorporate ocular
methods, aerial photographs, or satellite imagery at
one or more scales to estimate biomass, and then
measure biomass directly by harvesting or other
techniques from a smaller number of sub samples
of the same area (Bonham 1989; Heller and Ulliman 1983). These approaches can be extremely ef-