the absorption and backscattering coefficients, a, b b (see (16.3)). Due to their
additive nature, the absorption and backscattering coefficients are sums of products
of cross sections a
* and b b
* (see (16.4)) and the respective CPA concentrations. As
it was indicated in Chap. 16, the tabulated values of spectral cross-sections (or
otherwise the hydro-optical model) determined for Lake Ladoga will be used herein
for the simulations.
When processing remote sensing data collected over open ocean/sea waters,
most frequently used are the methods based on regression expressions relating the
chlorophyll concentration to the water-leaving radiance ratio at two wavelengths in
the blue and green spectral regions.
As well known, the chlorophyll absorption spectrum exhibits two major absorption
bands at 430–450 and 660–680 nm. When a water body containing phytoplankton
(referred to as chlorophyll) is illuminated by the natural light, water-leaving
radiance proves to be subdued (as compared to the radiance signal leaving clear
water, i.e. water devoid of chlorophyll) in the region 400–500 nm (l 1 ) and enhanced
at l > 580–600 nm. These changes result from, respectively, chlorophyll absorption and an increase in the number of scattering centers due to phytoplankton cells.
At the same time, in the upwelling radiance spectrum there is a region ~ 500–520
nm (l 2 ) where the signal remains nearly intact with increasing chlorophyll concentration. In this specific situation, there is a possibility to relate the normalized depth
of the “chlorophyll dip” (i.e. L u (þ0) at 430 nm) to the chlorophyll concentration:
Cchl ¼ a
L u l 1 ¼ 430 nm
ð
Þ
L u l 2 ¼ 520 nm
ð
Þ
Àb
;
(18.1)
where a and b are regression coefficients obtained from statistically ample data
incorporating concurrent in situ measurements of C chl and remotely sensed
L u (420 nm) and L u (520 nm).
A small but infinitesimal concentrations of dissolved organic carbon and
suspended minerals, the increase in chlorophyll content results in a displacement
of l 1 and l 2 to longer wavelengths [to 450–480 nm and 520, respectively]. In
these conditions, the above “band-ratio” approach remains applicable: various
l 1 /l 2 pairs (443/520), (443/550), (520/550), (520/670) could be tried for a given
combination of C chl , C sm , C doc to fit best Eq. 18.1. Indeed, such attempts proved to
be reasonably successful when dealing with clear off-shore ocean and marine
waters. However, as soon as suspended minerals and dissolved organics
concentrations sensibly increase (which is a characteristic of inland and marine
coastal waters), it becomes impossible to identify l 1 and l 2 in the spectral
distribution of the upwelling radiance, and the regression retrieval algorithm
becomes completely inefficient.
One of the modern algorithms for water quality retrieval is the method of
neural networks. It is based on approximation of the dependency of R(À0,l) on
combinations of CPA concentrations. In the first stage, a training array of data
containing the CPA concentration vector and the respective values of R(À0,l) at
176
18 Retrieval of CPA Concentrations from the Spectral Composition of Subsurface
additive nature, the absorption and backscattering coefficients are sums of products
of cross sections a
* and b b
* (see (16.4)) and the respective CPA concentrations. As
it was indicated in Chap. 16, the tabulated values of spectral cross-sections (or
otherwise the hydro-optical model) determined for Lake Ladoga will be used herein
for the simulations.
When processing remote sensing data collected over open ocean/sea waters,
most frequently used are the methods based on regression expressions relating the
chlorophyll concentration to the water-leaving radiance ratio at two wavelengths in
the blue and green spectral regions.
As well known, the chlorophyll absorption spectrum exhibits two major absorption
bands at 430–450 and 660–680 nm. When a water body containing phytoplankton
(referred to as chlorophyll) is illuminated by the natural light, water-leaving
radiance proves to be subdued (as compared to the radiance signal leaving clear
water, i.e. water devoid of chlorophyll) in the region 400–500 nm (l 1 ) and enhanced
at l > 580–600 nm. These changes result from, respectively, chlorophyll absorption and an increase in the number of scattering centers due to phytoplankton cells.
At the same time, in the upwelling radiance spectrum there is a region ~ 500–520
nm (l 2 ) where the signal remains nearly intact with increasing chlorophyll concentration. In this specific situation, there is a possibility to relate the normalized depth
of the “chlorophyll dip” (i.e. L u (þ0) at 430 nm) to the chlorophyll concentration:
Cchl ¼ a
L u l 1 ¼ 430 nm
ð
Þ
L u l 2 ¼ 520 nm
ð
Þ
Àb
;
(18.1)
where a and b are regression coefficients obtained from statistically ample data
incorporating concurrent in situ measurements of C chl and remotely sensed
L u (420 nm) and L u (520 nm).
A small but infinitesimal concentrations of dissolved organic carbon and
suspended minerals, the increase in chlorophyll content results in a displacement
of l 1 and l 2 to longer wavelengths [to 450–480 nm and 520, respectively]. In
these conditions, the above “band-ratio” approach remains applicable: various
l 1 /l 2 pairs (443/520), (443/550), (520/550), (520/670) could be tried for a given
combination of C chl , C sm , C doc to fit best Eq. 18.1. Indeed, such attempts proved to
be reasonably successful when dealing with clear off-shore ocean and marine
waters. However, as soon as suspended minerals and dissolved organics
concentrations sensibly increase (which is a characteristic of inland and marine
coastal waters), it becomes impossible to identify l 1 and l 2 in the spectral
distribution of the upwelling radiance, and the regression retrieval algorithm
becomes completely inefficient.
One of the modern algorithms for water quality retrieval is the method of
neural networks. It is based on approximation of the dependency of R(À0,l) on
combinations of CPA concentrations. In the first stage, a training array of data
containing the CPA concentration vector and the respective values of R(À0,l) at
176
18 Retrieval of CPA Concentrations from the Spectral Composition of Subsurface