Chapter 15 Remote Sensing of Seagrasses
355
further to:
R(0−, H ) = R ∞ + (R b − R ∞ )
× exp[−2K d H ]
(13)
which is identical to the formulation of Bierwirth
et al. (1993).
The above formulations form the basis for remote
sensing of a substratum covered by a water column.
As Chapter 13 illustrates, though, the complexity of
the seagrass canopy geometry due to leaf form and
movement with waves and currents, is not incorporated into these analytical derivations. A merger between these two analytical models (for water column
and seagrass canopy) is required for improved understanding of the effects of seagrass canopy structure
and variability of the remotely sensed signal.
V. Methodological Approaches
to Assessing Seagrass Ecosystem
Characteristics from Remote Sensing
What are the consequences of these formulations for
the remote sensing of seagrasses? Before we answer
this question, we need to discuss possible approaches
to remote sensing of seagrasses from the simple to
the more complex.
A. Choice of Methods
1. The Empirical Method
This is the favorite methodology for seagrass mapping and uses whatever remote sensing imagery
may be readily available at the study site using traditional supervised or unsupervised classification
procedures together with fieldwork to identify the
characteristic substratum types that appear in the
image including benthic plant species and other
cover classes. Fieldwork is preferably carried out
during the overpass of the remote sensor, but in
practice fieldwork is often performed days to weeks
before or after collection of the imagery. As the remote sensing image usually covers a much larger
area than surveyed during fieldwork, extrapolation
is performed using a variety of either subjective
(worst case) or statistically developed (best case)
techniques. No explicit use is made of spectral bands
or their positioning. Unfortunately there is no guarantee that the extrapolations are valid, nor can this
methodology deal with atmosphere, air–water interface, or water column depth differences or with substratums and vegetation covers that were not present
in the fieldwork. Another significant drawback is that
it is impossible to detect multitemporal change in an
objective manner, nor is it possible to transfer the image classes developed using this technique to other
areas or other images without having to repeat all
the fieldwork and analysis. Thus, this method relies
heavily on local expert knowledge.
2. The Semi-empirical
or Semi-analytical Method
With this approach, the results gained from using
a traditional classification procedure are improved
by applying some a priori knowledge of the spectral behavior of the substratum and vegetation cover
classes in consideration of the spectral bands available from the remote sensing instrument before image acquisition. An optimal band set is selected
for a specific mapping purpose. In the case of hyperspectral sensors or sensors with programmable
bands, the entire range of useful spectral bands may
be collected. Joyce and Phinn (2003) were able to
assess chlorophyll content and photosynthetic capacity of coral reef substratum using hyperspectral
measurements using a spectral matching approach.
Andrefouet et al. (2003) were able to discriminate
chlorophylls, carotenoids, and phycobilin pigments
over microbial mats on a coral atoll using airborne
imaging spectrometry data and by applying derivative analysis. Both the studies indicate that this will
become possible for seagrasses too.
3. The Analytical Method
With this method, the radiative transfer equations can
be simplified for the retrieval of the variables of interest, and hence, the weak benthic signal containing
the desired layer of information about the seagrasses
can be untangled from remote sensing image data.
For instance, Eq. (11) (or the more detailed equations
preceding that derivation) can be inverted for a spectral band to enable the determination of one variable
such as bottom reflectance (R b ) provided that all the
other variables are known or can be estimated. When
based solely on the remote sensing reflectance, only
(R(0−,H ) is measured (after atmospheric and air–
water interface correction).
355
further to:
R(0−, H ) = R ∞ + (R b − R ∞ )
× exp[−2K d H ]
(13)
which is identical to the formulation of Bierwirth
et al. (1993).
The above formulations form the basis for remote
sensing of a substratum covered by a water column.
As Chapter 13 illustrates, though, the complexity of
the seagrass canopy geometry due to leaf form and
movement with waves and currents, is not incorporated into these analytical derivations. A merger between these two analytical models (for water column
and seagrass canopy) is required for improved understanding of the effects of seagrass canopy structure
and variability of the remotely sensed signal.
V. Methodological Approaches
to Assessing Seagrass Ecosystem
Characteristics from Remote Sensing
What are the consequences of these formulations for
the remote sensing of seagrasses? Before we answer
this question, we need to discuss possible approaches
to remote sensing of seagrasses from the simple to
the more complex.
A. Choice of Methods
1. The Empirical Method
This is the favorite methodology for seagrass mapping and uses whatever remote sensing imagery
may be readily available at the study site using traditional supervised or unsupervised classification
procedures together with fieldwork to identify the
characteristic substratum types that appear in the
image including benthic plant species and other
cover classes. Fieldwork is preferably carried out
during the overpass of the remote sensor, but in
practice fieldwork is often performed days to weeks
before or after collection of the imagery. As the remote sensing image usually covers a much larger
area than surveyed during fieldwork, extrapolation
is performed using a variety of either subjective
(worst case) or statistically developed (best case)
techniques. No explicit use is made of spectral bands
or their positioning. Unfortunately there is no guarantee that the extrapolations are valid, nor can this
methodology deal with atmosphere, air–water interface, or water column depth differences or with substratums and vegetation covers that were not present
in the fieldwork. Another significant drawback is that
it is impossible to detect multitemporal change in an
objective manner, nor is it possible to transfer the image classes developed using this technique to other
areas or other images without having to repeat all
the fieldwork and analysis. Thus, this method relies
heavily on local expert knowledge.
2. The Semi-empirical
or Semi-analytical Method
With this approach, the results gained from using
a traditional classification procedure are improved
by applying some a priori knowledge of the spectral behavior of the substratum and vegetation cover
classes in consideration of the spectral bands available from the remote sensing instrument before image acquisition. An optimal band set is selected
for a specific mapping purpose. In the case of hyperspectral sensors or sensors with programmable
bands, the entire range of useful spectral bands may
be collected. Joyce and Phinn (2003) were able to
assess chlorophyll content and photosynthetic capacity of coral reef substratum using hyperspectral
measurements using a spectral matching approach.
Andrefouet et al. (2003) were able to discriminate
chlorophylls, carotenoids, and phycobilin pigments
over microbial mats on a coral atoll using airborne
imaging spectrometry data and by applying derivative analysis. Both the studies indicate that this will
become possible for seagrasses too.
3. The Analytical Method
With this method, the radiative transfer equations can
be simplified for the retrieval of the variables of interest, and hence, the weak benthic signal containing
the desired layer of information about the seagrasses
can be untangled from remote sensing image data.
For instance, Eq. (11) (or the more detailed equations
preceding that derivation) can be inverted for a spectral band to enable the determination of one variable
such as bottom reflectance (R b ) provided that all the
other variables are known or can be estimated. When
based solely on the remote sensing reflectance, only
(R(0−,H ) is measured (after atmospheric and air–
water interface correction).