Sun and Ranson 1995). HV backscatter is dominated by volume scattering from
the woody elements in the trees, so that HV is strongly related to AGB. For the HH
and VV polarisations, ground conditions can affect the biomass-backscatter relationship, because HH backscatter comes mainly from trunk-ground scattering
while VV backscatter results from both volume and ground scattering.
Forest biomass can be retrieved from radar data using regression models and
inversion models. Ranson and Sun (1994, 2000) and Saatchi et al. (2007a, b) used
nonlinear regression models to estimate AGB from polarimetric or dual-polarimetic
measurements (e.g., HV and HH) by simplifying the complex modeling formulation. At L-band frequency, the regression method can estimate AGB with a 20 %
precision up to 200 Mg/ha in boreal and temperate forests and 150 Mg/ha over
tropical forests (Saatchi et al. 2007a, b).
The inversion of scattering physical models is often categorized as either
random-media or structure-based methods. The random-media approach models
the canopy as layers of homogeneous random media (Treuhaft et al. 1996;
Treuhaft and Siqueira 2000). However, the structure-based method ‘‘grows’’
realizations of high fidelity, naturally-varying tree structures, and then examines
the scattering from a sample of trees to determine canopy scattering properties
(Sun and Ranson 1995). In a three-dimensional forest backscatter model, a ray
tracing method calculates backscattering components from crowns, trunk, and
ground, and scattering between crowns and back-ground, and between trunk and
ground (Sun and Ranson 1995).
Because the sensitivity of the backscattering coefficient to biomass decreases at
high levels of biomass, inversion methods based solely on intensity are insufficient
to cover the full range of the world’s biomass. To overcome this problem,
polarimetric interferometry is used to derive height of vegetation phase scattering
center, which is closely related to vegetation height characteristics. The interferometric coherence can be calculated using two images of a scene acquired at
different times (for a repeat pass system) and with slightly different geometries,
which is decomposed into the noise decorrelation, the temporal coherence, and the
volume decorrelation. The latter is effectively correlated to vegetation height.
A major advantage of this approach is that both height and biomass measurements
are provided independently by the same radar sensor. In addition, the sensitivity of
Polarimetric SAR Interferometry (Pol-InSAR) to height increases with height (and
hence biomass), whereas the sensitivity of intensity to biomass decreases with
biomass, so that the two measurements complement each other when used jointly
in retrieval (Treuhaft and Siqueira 2000).
Application of the radar biomass estimation at continental or globe scale is best
at 1.0 ha scale (100 9 100 m pixel size). At this scale, the distribution of AGB
over the landscape is both stationary and normal, so that the radar resolution is
large enough to reduce the speckle noise and the geolocation error between radar
pixel and the plot location. Errors associated with the biomass estimation from
radar backscatter or height measurements at this scale can be reduced to acceptable
levels (10–20 %) for mapping the aboveground biomass globally (Saatchi et al.
2011).
80
X. Zhang and W. Ni-meister
the woody elements in the trees, so that HV is strongly related to AGB. For the HH
and VV polarisations, ground conditions can affect the biomass-backscatter relationship, because HH backscatter comes mainly from trunk-ground scattering
while VV backscatter results from both volume and ground scattering.
Forest biomass can be retrieved from radar data using regression models and
inversion models. Ranson and Sun (1994, 2000) and Saatchi et al. (2007a, b) used
nonlinear regression models to estimate AGB from polarimetric or dual-polarimetic
measurements (e.g., HV and HH) by simplifying the complex modeling formulation. At L-band frequency, the regression method can estimate AGB with a 20 %
precision up to 200 Mg/ha in boreal and temperate forests and 150 Mg/ha over
tropical forests (Saatchi et al. 2007a, b).
The inversion of scattering physical models is often categorized as either
random-media or structure-based methods. The random-media approach models
the canopy as layers of homogeneous random media (Treuhaft et al. 1996;
Treuhaft and Siqueira 2000). However, the structure-based method ‘‘grows’’
realizations of high fidelity, naturally-varying tree structures, and then examines
the scattering from a sample of trees to determine canopy scattering properties
(Sun and Ranson 1995). In a three-dimensional forest backscatter model, a ray
tracing method calculates backscattering components from crowns, trunk, and
ground, and scattering between crowns and back-ground, and between trunk and
ground (Sun and Ranson 1995).
Because the sensitivity of the backscattering coefficient to biomass decreases at
high levels of biomass, inversion methods based solely on intensity are insufficient
to cover the full range of the world’s biomass. To overcome this problem,
polarimetric interferometry is used to derive height of vegetation phase scattering
center, which is closely related to vegetation height characteristics. The interferometric coherence can be calculated using two images of a scene acquired at
different times (for a repeat pass system) and with slightly different geometries,
which is decomposed into the noise decorrelation, the temporal coherence, and the
volume decorrelation. The latter is effectively correlated to vegetation height.
A major advantage of this approach is that both height and biomass measurements
are provided independently by the same radar sensor. In addition, the sensitivity of
Polarimetric SAR Interferometry (Pol-InSAR) to height increases with height (and
hence biomass), whereas the sensitivity of intensity to biomass decreases with
biomass, so that the two measurements complement each other when used jointly
in retrieval (Treuhaft and Siqueira 2000).
Application of the radar biomass estimation at continental or globe scale is best
at 1.0 ha scale (100 9 100 m pixel size). At this scale, the distribution of AGB
over the landscape is both stationary and normal, so that the radar resolution is
large enough to reduce the speckle noise and the geolocation error between radar
pixel and the plot location. Errors associated with the biomass estimation from
radar backscatter or height measurements at this scale can be reduced to acceptable
levels (10–20 %) for mapping the aboveground biomass globally (Saatchi et al.
2011).
80
X. Zhang and W. Ni-meister