3.6 Future Research Directions
The validation and accuracy analysis of satellite-derived forest biomass is one of
the most important and challenging tasks. The accurate estimation of forest
biomass is crucial for monitoring fuel wood availability, modeling global carbon
consequences, and managing forest change. Unfortunately, current biomass estimates of regional area are widely inconsistent. For example, seven products in
Uganda produce total AGB estimates that range from 343 to 2,201 Tg and also
present different spatial distribution patterns (Avitabile et al. 2011). Thus, it is
necessary to generate a set of reliable true measures of biomass from various
ecosystems and plant species, which are sufficient to account for spatial heterogeneity and to represent forest types. It is recognized that reconciliation of the
ground and satellite-based biomass is extremely challenging and that in situ
datasets collected across ecosystems at spatial scales commensurate with moderate-coarse resolution data are urgently required. To make substantial validation
possible in future, it is urgent to generate a series of core validation data sets by
up-scaling intensive field data or HRSI and lidar data to the Landsat pixel scale
and MODIS pixel scales.
In the mean time, biomass estimates from remote sensing currently still rely
heavily on field based training data sets. At regional and global scales, these field
measurements are always far from sufficient to represent complex forest properties. Thus, using physically-based or physical approaches to retrieve forest attributes and biomass seems the most promising and avenues of advancement.
However to accomplish this, more efforts are needed to investigate canopy
reflectance models and forest allometric models which can provide the possibility
to estimate biomass at continental scales with limited ground-based training
samples.
It must be acknowledged that forest biomass is a dynamic process governed by
disturbance and subsequent re-growth processes (Harmon et al. 1990; Wofsy and
Harris 2002; Kennedy et al. 2007). Currently biomass at two specific times
(images) is generally produced and the difference between the two different
periods (images) is commonly used to detect change (Coppin et al. 2004). The
detection of trajectory-based changes (Kennedy et al. 2007) is a more meaningful
method of monitoring forest biomass. Given that the global archive of long-term
Landsat data is being made available for free in a standard processing format
(Woodcock et al. 2008) and the time series of global MODIS and VIIRS data is
available for more than a decade, robust approaches are required to automatically
retrieve forest biomass trajectories.
Passive optical remote sensing, lidar, and microwave remote sensing have
advantages and disadvantages in forest biomass estimates. Fusion of both data
would be a promising tool. Lidar provides accurate measure of woody volume
while accurate estimates of AGB require vegetation types, which can be obtained
from passive optical remote sensing. The small footprint lidar data is only limited
to small regions. Large footprint lidar can directly measure the horizontal and
88
X. Zhang and W. Ni-meister
The validation and accuracy analysis of satellite-derived forest biomass is one of
the most important and challenging tasks. The accurate estimation of forest
biomass is crucial for monitoring fuel wood availability, modeling global carbon
consequences, and managing forest change. Unfortunately, current biomass estimates of regional area are widely inconsistent. For example, seven products in
Uganda produce total AGB estimates that range from 343 to 2,201 Tg and also
present different spatial distribution patterns (Avitabile et al. 2011). Thus, it is
necessary to generate a set of reliable true measures of biomass from various
ecosystems and plant species, which are sufficient to account for spatial heterogeneity and to represent forest types. It is recognized that reconciliation of the
ground and satellite-based biomass is extremely challenging and that in situ
datasets collected across ecosystems at spatial scales commensurate with moderate-coarse resolution data are urgently required. To make substantial validation
possible in future, it is urgent to generate a series of core validation data sets by
up-scaling intensive field data or HRSI and lidar data to the Landsat pixel scale
and MODIS pixel scales.
In the mean time, biomass estimates from remote sensing currently still rely
heavily on field based training data sets. At regional and global scales, these field
measurements are always far from sufficient to represent complex forest properties. Thus, using physically-based or physical approaches to retrieve forest attributes and biomass seems the most promising and avenues of advancement.
However to accomplish this, more efforts are needed to investigate canopy
reflectance models and forest allometric models which can provide the possibility
to estimate biomass at continental scales with limited ground-based training
samples.
It must be acknowledged that forest biomass is a dynamic process governed by
disturbance and subsequent re-growth processes (Harmon et al. 1990; Wofsy and
Harris 2002; Kennedy et al. 2007). Currently biomass at two specific times
(images) is generally produced and the difference between the two different
periods (images) is commonly used to detect change (Coppin et al. 2004). The
detection of trajectory-based changes (Kennedy et al. 2007) is a more meaningful
method of monitoring forest biomass. Given that the global archive of long-term
Landsat data is being made available for free in a standard processing format
(Woodcock et al. 2008) and the time series of global MODIS and VIIRS data is
available for more than a decade, robust approaches are required to automatically
retrieve forest biomass trajectories.
Passive optical remote sensing, lidar, and microwave remote sensing have
advantages and disadvantages in forest biomass estimates. Fusion of both data
would be a promising tool. Lidar provides accurate measure of woody volume
while accurate estimates of AGB require vegetation types, which can be obtained
from passive optical remote sensing. The small footprint lidar data is only limited
to small regions. Large footprint lidar can directly measure the horizontal and
88
X. Zhang and W. Ni-meister