applicable for matching with pixels of moderate-high resolution data, such as
HRSI and Landsat TM, at local regions.
In-situ data that are comparable with moderate resolution instrument (including
MODIS, AVHRR, MISR, and SPOT VGT) pixels are basically not available. As a
result, assessment instead of direct validation is implemented for continental
biomass estimates. Generally, biomass estimates from MODIS data are evaluated
using values estimated from lidar, panchromatic imagery of IKONOS and
QuickBird, Landsat TM data (Palace et al. 2008; Chopping 2010; Baccini et al.
2004; Zhang and Kondragunta 2006). To evaluate biomass estimates appropriately
at large scales, efforts are needed to upscale biomass measurements from plot (or
lidar measurements) to HRSI, Landsat TM and then to MODIS pixels.
Moreover, there are no standard approaches to validate biomass estimates from
satellites. The commonly used indices for validation/assessment of biomass are:
correlation of determination (R
2 ), total biomass, mean biomass, root mean square
error (RMSE), relative RMSE, bias, and relative bias (Heiskanen 2006a;
Labrecque et al. 2006; Powell et al. 2010). High quality of biomass results demonstrated from one index is not necessary to correspond to high quality based on
other indices. In a given area, therefore, high quality of biomass estimates needs to
be qualified by considering various indices described above.
3.5 Major Findings
Selecting an optimal method from numerous models and satellite instruments for
biomass calculations needs to follow several principles. Broadly, methods are
acceptable if they meet their objectives or design requirements. Simply speaking,
the reasonableness of the methods and the availability of data are the main principles when selecting a method. All biomass models seek to simplify the complexity of forest properties by selectively exaggerating the fundamental variables.
Several simple rules are suggested to estimate forest biomass from satellite data.
First, a simplest method that will provide acceptable accuracy should then be
adopted. Secondly, the assumptions and limitations of the method (model) should
always be remembered, and the degree of uncertainty associated with model
predictions should always be known. More complex variables used in a model are
clearly more versatile, but such method may also be difficult to use widely.
Because the presence or absence of available input data constraints the method
selection, the availability of data will determine which method may be selected.
Numerous regression modeling approaches have been proposed for empirical
estimation of aboveground biomass with satellite variables and biophysical data. A
regression model is commonly created by selecting samples in field biomass and
satellite variables from one single scene, such as Landsat data. The model is then
applied to the entire scene, which is expected to work well in the given condition
and satellite scene. However, it is impracticable to directly transfer the model
across biomes and the time periods of satellite data because spectral properties
3 Remote Sensing of Forest Biomass
85
HRSI and Landsat TM, at local regions.
In-situ data that are comparable with moderate resolution instrument (including
MODIS, AVHRR, MISR, and SPOT VGT) pixels are basically not available. As a
result, assessment instead of direct validation is implemented for continental
biomass estimates. Generally, biomass estimates from MODIS data are evaluated
using values estimated from lidar, panchromatic imagery of IKONOS and
QuickBird, Landsat TM data (Palace et al. 2008; Chopping 2010; Baccini et al.
2004; Zhang and Kondragunta 2006). To evaluate biomass estimates appropriately
at large scales, efforts are needed to upscale biomass measurements from plot (or
lidar measurements) to HRSI, Landsat TM and then to MODIS pixels.
Moreover, there are no standard approaches to validate biomass estimates from
satellites. The commonly used indices for validation/assessment of biomass are:
correlation of determination (R
2 ), total biomass, mean biomass, root mean square
error (RMSE), relative RMSE, bias, and relative bias (Heiskanen 2006a;
Labrecque et al. 2006; Powell et al. 2010). High quality of biomass results demonstrated from one index is not necessary to correspond to high quality based on
other indices. In a given area, therefore, high quality of biomass estimates needs to
be qualified by considering various indices described above.
3.5 Major Findings
Selecting an optimal method from numerous models and satellite instruments for
biomass calculations needs to follow several principles. Broadly, methods are
acceptable if they meet their objectives or design requirements. Simply speaking,
the reasonableness of the methods and the availability of data are the main principles when selecting a method. All biomass models seek to simplify the complexity of forest properties by selectively exaggerating the fundamental variables.
Several simple rules are suggested to estimate forest biomass from satellite data.
First, a simplest method that will provide acceptable accuracy should then be
adopted. Secondly, the assumptions and limitations of the method (model) should
always be remembered, and the degree of uncertainty associated with model
predictions should always be known. More complex variables used in a model are
clearly more versatile, but such method may also be difficult to use widely.
Because the presence or absence of available input data constraints the method
selection, the availability of data will determine which method may be selected.
Numerous regression modeling approaches have been proposed for empirical
estimation of aboveground biomass with satellite variables and biophysical data. A
regression model is commonly created by selecting samples in field biomass and
satellite variables from one single scene, such as Landsat data. The model is then
applied to the entire scene, which is expected to work well in the given condition
and satellite scene. However, it is impracticable to directly transfer the model
across biomes and the time periods of satellite data because spectral properties
3 Remote Sensing of Forest Biomass
85