field plots were paired to the nearest pixel possible in a 2.4-m-resolution QuickBird
image for chlorophyll content estimation at the canopy (0.5 m ´ 0.5 m) and landscape
(50 m ´ 50 m); therefore, a slight displacement in georeferencing may still exist. In
spite of these differences, the significant results at the canopy and landscape scale
imply that it is possible to accurately estimate chlorophyll contents using both ground
hyperspectral and satellite multispectral remote sensing data. The recent availability
of spaceborne hyperspectral data such as CHRIS/PROBA (the Compact High
Resolution Imaging Spectrometer/Project for On Board Autonomy) and the development of more effective narrow-band chlorophyll indices provide new opportunity
to estimate the biochemical properties of vegetation at the landscape or even
higher level.
As scales increased from leaf to landscape, more and more new spectral features
are included in the remote sensing data. For example, at the leaf scale, leaf Chl a + b
variation and anatomical differences such as cell size, air spaces, and cell wall
thickness can all contribute to spectral differences. At the canopy level, besides leaf
characteristics, canopy compositions including standing dead, litter, and bare soil are
incorporated into the spectral data. At the landscape level, a complex array of ground
and atmospheric features and sensing techniques, including illumination and viewing
geometry, affect the imagery data. One would thus assume that relationships between
vegetation chlorophyll properties and spectral indices would become weaker and
weaker as scale increased from species to landscape level. This is true when the
chlorophyll is scaled from the leaf to the canopy level, and the R
2 value is decreased
from 64% of leaf level to 53% of canopy level. Unexpectedly, this study found that the
relationships at the landscape level became stronger than that in the landscape and
level levels. The unexpected result may be explained by a few factors: (1) The less
regression samples (only 16 sites in the landscape level) with large variation in the
sampling sites compared to much more regression samples at the canopy levels with
less variation in the samples. (2) When scaling the Chl a + b data from leaf to
canopy level, the variation in species percentage cover may contribute to higher
variation in canopy Chl a + b data. Further, when scaling the Chl a + b data from
the canopy to the landscape level, an effect of averaging could potentially exist.
Cumulatively, these factors probably contributed to the higher coefficients of
determinations at the landscape level. These factors also represent inherent
limitations to using the approach as a scaling tool, unless the sampling points at
three scales are controlled and variation in the samples is taken into account in
future studies and applications.
Chlorophyll content in leaf keeps changing through the entire growing season,
while our analysis is only based on one set of field data from July 1 to July 15, 2011.
The regression models established in this study may only be suitable for estimating
chlorophyll content in the maximum growing season. In the early or late growing
season, other factors in the field could cause a different relationship between
chlorophyll content and spectral indices. A time-series analysis is thus needed to
provide more insights on the relationships between chlorophyll content and spectral
indices over the growing season.
CONCLUSIONS AND DISCUSSION
135
image for chlorophyll content estimation at the canopy (0.5 m ´ 0.5 m) and landscape
(50 m ´ 50 m); therefore, a slight displacement in georeferencing may still exist. In
spite of these differences, the significant results at the canopy and landscape scale
imply that it is possible to accurately estimate chlorophyll contents using both ground
hyperspectral and satellite multispectral remote sensing data. The recent availability
of spaceborne hyperspectral data such as CHRIS/PROBA (the Compact High
Resolution Imaging Spectrometer/Project for On Board Autonomy) and the development of more effective narrow-band chlorophyll indices provide new opportunity
to estimate the biochemical properties of vegetation at the landscape or even
higher level.
As scales increased from leaf to landscape, more and more new spectral features
are included in the remote sensing data. For example, at the leaf scale, leaf Chl a + b
variation and anatomical differences such as cell size, air spaces, and cell wall
thickness can all contribute to spectral differences. At the canopy level, besides leaf
characteristics, canopy compositions including standing dead, litter, and bare soil are
incorporated into the spectral data. At the landscape level, a complex array of ground
and atmospheric features and sensing techniques, including illumination and viewing
geometry, affect the imagery data. One would thus assume that relationships between
vegetation chlorophyll properties and spectral indices would become weaker and
weaker as scale increased from species to landscape level. This is true when the
chlorophyll is scaled from the leaf to the canopy level, and the R
2 value is decreased
from 64% of leaf level to 53% of canopy level. Unexpectedly, this study found that the
relationships at the landscape level became stronger than that in the landscape and
level levels. The unexpected result may be explained by a few factors: (1) The less
regression samples (only 16 sites in the landscape level) with large variation in the
sampling sites compared to much more regression samples at the canopy levels with
less variation in the samples. (2) When scaling the Chl a + b data from leaf to
canopy level, the variation in species percentage cover may contribute to higher
variation in canopy Chl a + b data. Further, when scaling the Chl a + b data from
the canopy to the landscape level, an effect of averaging could potentially exist.
Cumulatively, these factors probably contributed to the higher coefficients of
determinations at the landscape level. These factors also represent inherent
limitations to using the approach as a scaling tool, unless the sampling points at
three scales are controlled and variation in the samples is taken into account in
future studies and applications.
Chlorophyll content in leaf keeps changing through the entire growing season,
while our analysis is only based on one set of field data from July 1 to July 15, 2011.
The regression models established in this study may only be suitable for estimating
chlorophyll content in the maximum growing season. In the early or late growing
season, other factors in the field could cause a different relationship between
chlorophyll content and spectral indices. A time-series analysis is thus needed to
provide more insights on the relationships between chlorophyll content and spectral
indices over the growing season.
CONCLUSIONS AND DISCUSSION
135