Vertucci and Likens (1989) studied a group of 44 low to moderate productivity
Adirondack mountain lakes in which the non-acid corrected chl a averages for five lake
classes ranged from 0.3 - 4.8 µg/l. These lakes ranged from Case 1, low productivity,
blue water lakes to Case 2 conditions with moderate chl a, variable CDOM, and colors
ranging from gray-brown to yellow-green to brown. Optimization procedures were used
to determine the best wavelength combinations within visible and lower NIR regions.
In these lakes, a two band ratio of 525 nm to 554 nm (Table 1) performed significantly
better than the band ratio of 443 to 550 nm. A bio-optical model using pigment, seston,
and CDOM data produced model spectra very similar to their field spectra.
A majority of the semi-empirical algorithms for chlorophyll estimation in Case 2
waters utilize some variant of NIR peak height to chl a red absorption (Table 1). In a
number of these studies, a simple two band ratio of the NIR peak, at either a fixed
wavelength or at its maximum value is normalized to the minimum of reflectance at or
near 675 nm. Kallio (2003) used a first order, linear equation with a band ratio of 705 to
662 nm to predict chlorophyll from AISA imagery data in two Finnish lakes with a chl
a range of 6 to 70 µg/l. Thiemann and Kaufmann (2000) used a similar band ratio of
705 to 678 nm ratio to calibrate aerial hyperspectral imagery of lakes in northern
Germany (chl a range of 5 to 350 µg/l). Mittenzwey et al. (1992) fit a second order
polynomial to a band ratio of 705 to 670 nm in a set of lakes and rivers in Germany
with a much wider chl a range of 5 - 350 µg/l (Table 1).
As discussed previously, Schalles et al. (1998b) and Gitelson et al. (2000)
determined that a model using the height of the NIR peak above a normalizing baseline
from 675 to 750 nm (Figure 26) was the best algorithm for eutrophic midwestern lakes
in the United States and for Lake Kinneret and Haifa Bay in Israel. Gons (1999)
modified the simple two band ratio (704 to 672 nm) approach using a bio-optical
model. His approach required an inversion model for reflectance at 776 nm to estimate
the backscatter coefficient b b . The backscatter term became an adjustment for extra
scattering by tripton particles, when combined with the respective water absorption
coefficients for the 672 and 704 nm bands. This approach provided a robust chlorophyll
prediction model for a diverse set of inland and coastal Case 2 waters with a chl a range
of 2 to 994 µg/l. Gons et al. (2002) and Ruddick et al. (2004) modified this bio-optical
model to utilize available bands on the MERIS satellite sensor. Dall ’Olmo et al. (2003)
proposed a novel technique, with an algorithm derived from pigment estimation in
terrestrial vegetation (Table 1). Their model was parameterized with Case 2 data from
turbid and productive lakes and reservoirs in Nebraska, USA with a chl a range of 7 to
194 and a seston range of < 0.1 to 214 mg/l. Their model, as with Gons et al. (2002),
uses a NIR band (740 to 750 nm) to compensate for backscatter by non-algal particles.
However, their model uses a difference in the reciprocals of two bands (660 to 670 nm
and 720 to 730 nm) to quantify pigment specific absorption activity and doesn’t utilize
the contrast in the NIR peak and chl a red trough features.
My laboratory recently participated in remote sensing studies at five National
Estuarine Research Reserves. We examined water reflectance patterns within estuarine
longitudinal mixing gradients and across watersheds and coastal provinces (Hladik,
2004; Schalles and Hladik, 2004). Optical measurements and analysis of OACs were
made at Apalachicola Bay, Florida (October, 2002), Sapelo Island, Georgia and vicinity
(January and August, 2003 and June, 2004), ACE Basin, South Carolina (June, 2003),
Grand Bay, Mississippi (October, 2003), and Delaware Bay near Dover, Delaware
(July, 2004). Reflectance was estimated from upwelling and downwelling signals
collected simultaneously with a pair of fiber optic cables connected to separate
68
Schalles
Adirondack mountain lakes in which the non-acid corrected chl a averages for five lake
classes ranged from 0.3 - 4.8 µg/l. These lakes ranged from Case 1, low productivity,
blue water lakes to Case 2 conditions with moderate chl a, variable CDOM, and colors
ranging from gray-brown to yellow-green to brown. Optimization procedures were used
to determine the best wavelength combinations within visible and lower NIR regions.
In these lakes, a two band ratio of 525 nm to 554 nm (Table 1) performed significantly
better than the band ratio of 443 to 550 nm. A bio-optical model using pigment, seston,
and CDOM data produced model spectra very similar to their field spectra.
A majority of the semi-empirical algorithms for chlorophyll estimation in Case 2
waters utilize some variant of NIR peak height to chl a red absorption (Table 1). In a
number of these studies, a simple two band ratio of the NIR peak, at either a fixed
wavelength or at its maximum value is normalized to the minimum of reflectance at or
near 675 nm. Kallio (2003) used a first order, linear equation with a band ratio of 705 to
662 nm to predict chlorophyll from AISA imagery data in two Finnish lakes with a chl
a range of 6 to 70 µg/l. Thiemann and Kaufmann (2000) used a similar band ratio of
705 to 678 nm ratio to calibrate aerial hyperspectral imagery of lakes in northern
Germany (chl a range of 5 to 350 µg/l). Mittenzwey et al. (1992) fit a second order
polynomial to a band ratio of 705 to 670 nm in a set of lakes and rivers in Germany
with a much wider chl a range of 5 - 350 µg/l (Table 1).
As discussed previously, Schalles et al. (1998b) and Gitelson et al. (2000)
determined that a model using the height of the NIR peak above a normalizing baseline
from 675 to 750 nm (Figure 26) was the best algorithm for eutrophic midwestern lakes
in the United States and for Lake Kinneret and Haifa Bay in Israel. Gons (1999)
modified the simple two band ratio (704 to 672 nm) approach using a bio-optical
model. His approach required an inversion model for reflectance at 776 nm to estimate
the backscatter coefficient b b . The backscatter term became an adjustment for extra
scattering by tripton particles, when combined with the respective water absorption
coefficients for the 672 and 704 nm bands. This approach provided a robust chlorophyll
prediction model for a diverse set of inland and coastal Case 2 waters with a chl a range
of 2 to 994 µg/l. Gons et al. (2002) and Ruddick et al. (2004) modified this bio-optical
model to utilize available bands on the MERIS satellite sensor. Dall ’Olmo et al. (2003)
proposed a novel technique, with an algorithm derived from pigment estimation in
terrestrial vegetation (Table 1). Their model was parameterized with Case 2 data from
turbid and productive lakes and reservoirs in Nebraska, USA with a chl a range of 7 to
194 and a seston range of < 0.1 to 214 mg/l. Their model, as with Gons et al. (2002),
uses a NIR band (740 to 750 nm) to compensate for backscatter by non-algal particles.
However, their model uses a difference in the reciprocals of two bands (660 to 670 nm
and 720 to 730 nm) to quantify pigment specific absorption activity and doesn’t utilize
the contrast in the NIR peak and chl a red trough features.
My laboratory recently participated in remote sensing studies at five National
Estuarine Research Reserves. We examined water reflectance patterns within estuarine
longitudinal mixing gradients and across watersheds and coastal provinces (Hladik,
2004; Schalles and Hladik, 2004). Optical measurements and analysis of OACs were
made at Apalachicola Bay, Florida (October, 2002), Sapelo Island, Georgia and vicinity
(January and August, 2003 and June, 2004), ACE Basin, South Carolina (June, 2003),
Grand Bay, Mississippi (October, 2003), and Delaware Bay near Dover, Delaware
(July, 2004). Reflectance was estimated from upwelling and downwelling signals
collected simultaneously with a pair of fiber optic cables connected to separate
68
Schalles