24
fective as long as there is moderate to high correlation between the indirect estimates and direct
biomass measures (The theory of multistage sampling is presented in detail by Brewer and Hanif
[1983].)
Ratio-based indices, linear models, and geometrical optical models have been used to estimate a
variety of canopy biophysical properties directly
linked to primary productivity including LAI, intercepted PAR, and phytomass (e.g., Tucker 1979;
Asrar et aI. 1989; see also Chapter 3). Indices, such
as the NDVI and spectral ratio (SR), and linear
techniques, such as the KT transform and PVI, have
been used to estimate green biomass primarily
through empirical relationships developed for crop
plants and grassland ecosystems. Alternative approaches include the use of process-based models
that link vegetation indices to meteorological variables, canopy aerodynamic resistance, net radiation, and physiological variables to compute seasonal changes in aboveground biomass (e.g.,
Running and Coughlan 1988, Asrar et al., 1989).
These approaches have been highly successful for
agriculture and grassland ecosystems and somewhat less successful in forested ecosystems (Hall et
aI., 1995). Myneni et aI. (1995) provide a current
review of models and algorithms used to derive
vegetation structural information from optical
imagery.
Geometrical optical models offer an alternative
approach toward estimating green biomass and timber volume in woodland and forested ecosystems.
For example, Wu and Strahler (1994) inverted an
optical geometrical model to estimate tree crown
radius and stand density and then used allometric
equations to predict foliar, woody, and total biomass for nine conifer forest stands. Estimates of
foliar biomass were highly correlated with direct
estimates based on measured basal area (,2
= 0.75). Indirect and direct methods were not as
strongly correlated for total biomass (,2 = 0.63).
Hall et aI. (1995) linked linear spectral mixture
analysis with a cylindrical canopy geometrical
model to model reflectance of boreal forests as a
product of sunlit and shadowed canopy fractions.
Using this approach, they were able to estimate biomass, density, average diameter at breast height
(DBH), LAI, and aboveground NPP. In comparison, they found the NDVI was a poor predictor of
canopy biophysical properties in these forests.
Frank W. Davis and Dar Roberts
Because optical indices saturate at low levels of
biomass, much attention has focused on the use of
radar imagery for mapping and monitoring biomass
in shrubland and forest ecosystems. Radar profilers
look especially promising because they can accurately retrieve forest height. Kasischke et aI. (1997)
cite numerous studies that have demonstrated the
sensitivity of imaging radar backscatter to woody
plant biomass. Radar backscatter saturates at low
to moderate levels of biomass (10 to 20 kg m - 2)
and is highly dependent on radar wavelength and
polarization, with the saturation level being higher
for longer wavelengths. Among single band systems, P-HV and L-HV have proven the most useful
wavelengths and polarizations.
Using multiband imagery and multistep estimation procedures, it may be possible to overcome
saturation effects and obtain quite good estimates
of forest biomass over large regions. For example,
Dobson et aI. (1995) first segmented the region into
more uniform forest structural types and then used
different wavelengths and polarizations to obtain
separate estimates of canopy biomass, canopy
height, and basal area, and then combined these to
estimate total forest biomass in mixed coniferousdeciduous stands of northern Michigan. They
achieved a correlation (,2) of 0.95 between SARestimated and allometric ally estimated biomass
without any obvious saturation effect. Other studies
have not attained such high accuracies and have not
overcome saturation effects beyond 15 to 20
kg m -2 (Ranson and Sun 1997, Harrell et al. 1997).
This could be due in part to the effects of other
scene-dependent factors, such as soil moisture and
surface roughness. All such studies are also subject
to significant uncertainties in ground-based estimates of total biomass for the test stands.
Three-Dimensional Structure
Direct Methods
Nearly all work to reconstruct three-dimensional
structure directly has been conducted in crops and
other herbaceous vegetation and has emphasized
the architecture of foliar elements. At a minimum,
measures of leaf height, area, and orientation (normal to the leaf surface) are needed to reconstruct
three-dimensional foliar structure. Stem structure
requires measures of stem length, circumference,
fective as long as there is moderate to high correlation between the indirect estimates and direct
biomass measures (The theory of multistage sampling is presented in detail by Brewer and Hanif
[1983].)
Ratio-based indices, linear models, and geometrical optical models have been used to estimate a
variety of canopy biophysical properties directly
linked to primary productivity including LAI, intercepted PAR, and phytomass (e.g., Tucker 1979;
Asrar et aI. 1989; see also Chapter 3). Indices, such
as the NDVI and spectral ratio (SR), and linear
techniques, such as the KT transform and PVI, have
been used to estimate green biomass primarily
through empirical relationships developed for crop
plants and grassland ecosystems. Alternative approaches include the use of process-based models
that link vegetation indices to meteorological variables, canopy aerodynamic resistance, net radiation, and physiological variables to compute seasonal changes in aboveground biomass (e.g.,
Running and Coughlan 1988, Asrar et al., 1989).
These approaches have been highly successful for
agriculture and grassland ecosystems and somewhat less successful in forested ecosystems (Hall et
aI., 1995). Myneni et aI. (1995) provide a current
review of models and algorithms used to derive
vegetation structural information from optical
imagery.
Geometrical optical models offer an alternative
approach toward estimating green biomass and timber volume in woodland and forested ecosystems.
For example, Wu and Strahler (1994) inverted an
optical geometrical model to estimate tree crown
radius and stand density and then used allometric
equations to predict foliar, woody, and total biomass for nine conifer forest stands. Estimates of
foliar biomass were highly correlated with direct
estimates based on measured basal area (,2
= 0.75). Indirect and direct methods were not as
strongly correlated for total biomass (,2 = 0.63).
Hall et aI. (1995) linked linear spectral mixture
analysis with a cylindrical canopy geometrical
model to model reflectance of boreal forests as a
product of sunlit and shadowed canopy fractions.
Using this approach, they were able to estimate biomass, density, average diameter at breast height
(DBH), LAI, and aboveground NPP. In comparison, they found the NDVI was a poor predictor of
canopy biophysical properties in these forests.
Frank W. Davis and Dar Roberts
Because optical indices saturate at low levels of
biomass, much attention has focused on the use of
radar imagery for mapping and monitoring biomass
in shrubland and forest ecosystems. Radar profilers
look especially promising because they can accurately retrieve forest height. Kasischke et aI. (1997)
cite numerous studies that have demonstrated the
sensitivity of imaging radar backscatter to woody
plant biomass. Radar backscatter saturates at low
to moderate levels of biomass (10 to 20 kg m - 2)
and is highly dependent on radar wavelength and
polarization, with the saturation level being higher
for longer wavelengths. Among single band systems, P-HV and L-HV have proven the most useful
wavelengths and polarizations.
Using multiband imagery and multistep estimation procedures, it may be possible to overcome
saturation effects and obtain quite good estimates
of forest biomass over large regions. For example,
Dobson et aI. (1995) first segmented the region into
more uniform forest structural types and then used
different wavelengths and polarizations to obtain
separate estimates of canopy biomass, canopy
height, and basal area, and then combined these to
estimate total forest biomass in mixed coniferousdeciduous stands of northern Michigan. They
achieved a correlation (,2) of 0.95 between SARestimated and allometric ally estimated biomass
without any obvious saturation effect. Other studies
have not attained such high accuracies and have not
overcome saturation effects beyond 15 to 20
kg m -2 (Ranson and Sun 1997, Harrell et al. 1997).
This could be due in part to the effects of other
scene-dependent factors, such as soil moisture and
surface roughness. All such studies are also subject
to significant uncertainties in ground-based estimates of total biomass for the test stands.
Three-Dimensional Structure
Direct Methods
Nearly all work to reconstruct three-dimensional
structure directly has been conducted in crops and
other herbaceous vegetation and has emphasized
the architecture of foliar elements. At a minimum,
measures of leaf height, area, and orientation (normal to the leaf surface) are needed to reconstruct
three-dimensional foliar structure. Stem structure
requires measures of stem length, circumference,