Similarly, other empirical approaches, such as neural networks (e.g., Schiller and
Doerffer 1999; Dzwonkowski and Yan 2005; Schroeder et al. 2007) that require
algorithm tuning often use regional data, and are thus applicable to regions with
similar optical properties. On the other hand, empirical algorithms are relatively
easy to implement, and the resulting satellite images have smooth transitions
between different regions when the same algorithm coefficients are used. More
importantly, some of the residual errors from atmospheric correction, often spectrally related, are partially removed in the band-ratio OCx algorithms and nearly
completely removed in the band-subtraction CI algorithm.
Empirical algorithms do not separate Chl from other OSCs, but treat all OSCs
as a whole. In contrast, semi-analytical algorithms derive all OSCs (including Chl)
simultaneously, or derive a– ph first and then use a pre-defined Chl-a ph relationship
to derive Chl (e.g., Sathyendranath et al. 1989; Roesler and Perry 1995; Hoge and
Lyon 1996; Carder et al. 1991, 1999; Lee et al. 1999, 2002; Maritorena et al.
2002). In these approaches, the absorption spectral shapes of CDOM, detritus, and
phytoplankton pigments are often derived from global datasets and assumed timeand space-independent. Therefore, in waters (especially coastal waters) where
these absorption shapes differ significantly from the global mean, algorithm
retuning is required. Because the emphasis of this chapter is on empirical algorithms, interested readers are referred to the published literature to get more indepth knowledge on semi-analytical algorithms.
Coastal waters, especially river plumes and estuaries, often have a significant
amount of CDOM from terrestrial discharge that dominates the light absorption in
the blue, resulting in a poor relationship between blue/green band ratios and Chl
(e.g., Odriozola et al. 2007). Figure 7.5a shows an example of the poor performance of blue/green band ratio algorithms (OC3 and OC4) in deriving Chl for a
moderately turbid estuary, Tampa Bay (Le et al. 2013). Under these circumstances,
spectral bands in the red and NIR, which are less affected by CDOM than in the
blue, can be used to avoid this problem. MODIS and MERIS are equipped with
bands specifically designed to quantify solar-stimulated phytoplankton fluorescence in the red, and band-subtraction algorithms have been developed to derive
fluorescence line height (FLH, Letilier et al. 1996) and maximum chlorophyll
index (MCI, Gower et al. 2005) as proxies for Chl. Application of the MODIS
FLH over the global open ocean, after adjustments for non-photochemical
quenching and phytoplankton packaging effects, showed excellent agreement with
the band-ratio-derived Chl for a large dynamic range (Behrenfeld et al. 2009).
Application in SW Florida coastal waters also showed tight correlation with in situ
Chl in CDOM-rich waters (Hu et al. 2005). Moreover, radiative transfer simulations showed that FLH is insensitive to CDOM changes, such that a 10-fold
increase in CDOM only resulted in a 50 % decrease in FLH (McKee et al. 2007b).
However, for sediment-rich waters, FLH is positively biased due to the unequal
contribution of the sediments to the reflectance in the FLH bands (Gilerson et al.
2007), resulting in a poor relationship between FLH and Chl.
Other forms of empirical algorithms using various band combinations in the red
and NIR have been proposed to avoid the CDOM contamination problems and to
182
C. Hu and J. Campbell
Doerffer 1999; Dzwonkowski and Yan 2005; Schroeder et al. 2007) that require
algorithm tuning often use regional data, and are thus applicable to regions with
similar optical properties. On the other hand, empirical algorithms are relatively
easy to implement, and the resulting satellite images have smooth transitions
between different regions when the same algorithm coefficients are used. More
importantly, some of the residual errors from atmospheric correction, often spectrally related, are partially removed in the band-ratio OCx algorithms and nearly
completely removed in the band-subtraction CI algorithm.
Empirical algorithms do not separate Chl from other OSCs, but treat all OSCs
as a whole. In contrast, semi-analytical algorithms derive all OSCs (including Chl)
simultaneously, or derive a– ph first and then use a pre-defined Chl-a ph relationship
to derive Chl (e.g., Sathyendranath et al. 1989; Roesler and Perry 1995; Hoge and
Lyon 1996; Carder et al. 1991, 1999; Lee et al. 1999, 2002; Maritorena et al.
2002). In these approaches, the absorption spectral shapes of CDOM, detritus, and
phytoplankton pigments are often derived from global datasets and assumed timeand space-independent. Therefore, in waters (especially coastal waters) where
these absorption shapes differ significantly from the global mean, algorithm
retuning is required. Because the emphasis of this chapter is on empirical algorithms, interested readers are referred to the published literature to get more indepth knowledge on semi-analytical algorithms.
Coastal waters, especially river plumes and estuaries, often have a significant
amount of CDOM from terrestrial discharge that dominates the light absorption in
the blue, resulting in a poor relationship between blue/green band ratios and Chl
(e.g., Odriozola et al. 2007). Figure 7.5a shows an example of the poor performance of blue/green band ratio algorithms (OC3 and OC4) in deriving Chl for a
moderately turbid estuary, Tampa Bay (Le et al. 2013). Under these circumstances,
spectral bands in the red and NIR, which are less affected by CDOM than in the
blue, can be used to avoid this problem. MODIS and MERIS are equipped with
bands specifically designed to quantify solar-stimulated phytoplankton fluorescence in the red, and band-subtraction algorithms have been developed to derive
fluorescence line height (FLH, Letilier et al. 1996) and maximum chlorophyll
index (MCI, Gower et al. 2005) as proxies for Chl. Application of the MODIS
FLH over the global open ocean, after adjustments for non-photochemical
quenching and phytoplankton packaging effects, showed excellent agreement with
the band-ratio-derived Chl for a large dynamic range (Behrenfeld et al. 2009).
Application in SW Florida coastal waters also showed tight correlation with in situ
Chl in CDOM-rich waters (Hu et al. 2005). Moreover, radiative transfer simulations showed that FLH is insensitive to CDOM changes, such that a 10-fold
increase in CDOM only resulted in a 50 % decrease in FLH (McKee et al. 2007b).
However, for sediment-rich waters, FLH is positively biased due to the unequal
contribution of the sediments to the reflectance in the FLH bands (Gilerson et al.
2007), resulting in a poor relationship between FLH and Chl.
Other forms of empirical algorithms using various band combinations in the red
and NIR have been proposed to avoid the CDOM contamination problems and to
182
C. Hu and J. Campbell