11. Inverse Models to Relate Inherent Optical Properties and Chlorophyll Using
Reflectance Spectra
The problems with site and temporal specificity in the above approaches using
empirical and semi-analytical algorithms have motivated numerous researchers to
develop analytical model approaches to link IOPs with measured AOPs, including
reflectance spectra (Doerffer and Fischer, 1991; Dekker et al., 1997; Lahet et al., 2001;
Sathyendranath et al., 2001; Pozdnyakov et al., 2002; Brando and Dekker, 2003;
Bowers et al., 2004; Gege and Albert, this volume). If IOPs are well characterized and
field reflectance data are carefully collected, inverse models can accurately retrieve
water constituents (Bukata et al., 1995). However, the inverse approach is highly
sensitive to errors in the measured radiometric variables, and even small errors may
invalidate the inversions (Mobley, 1994).
An obvious problem is finding unique solutions for inversion problems in which
optical water constituents and boundary conditions vary. Potentially, different sets of
boundary and IOP conditions may lead to the same solution. Iterative, Monte Carlo
techniques are typically used to find an optimal solution. Boundary conditions
(atmospheric aerosols, water surface conditions, sun angle) may be very dynamic,
particularly in coastal areas. High accuracy in atmospheric correction is crucial to
success in discriminating water constituents, especially with satellite data (Brando and
Dekker, 2003). The ability to differentiate optically shallow versus optically deep
waters and the contribution of bottom reflectance is also important. Lee et al. (1999)
developed a model which can derive bottom depths and water column properties from
reflectance spectra. Heterogeneity in the IOPs is a further challenge. For example,
differences in the slope of the absorption spectra of CDOM and detrital tripton
(Bricaud et al., 1991; Carder et al., 1989; Dekker, 1993; Bowers et al., 2003),
variability in pigment composition and cell packaging effects, and differences in
volume scattering functions of the different classes of sestonic particles (Dekker et al.,
1997) can all reduce the robustness of the assumed IOPs in coastal settings. Thus, it
becomes necessary to measure and define specific IOPs (SIOPs) for a given site, and
perhaps for a specific time (Brando and Dekker, 2003). Many of the inverse procedures
reduce the optically active constituents to the three primary classes reviewed here (algal
pigment, seston, and CDOM). However, these classes are often, themselves, complex
mixes of constituents. Tidal fluxes, river plumes, and other physical mechanisms,
which create spatial and temporal variability, can complicate the application of
inversion models to large areas measured with satellite sensors (Doerffer et al., 1994).
In many cases, it may not be practical for ship based researchers to synoptically
measure water properties for IOP estimation. In fairness, this set of considerations often
applies to the simpler, band ratio algorithms discussed above.
In spite of the serious constraints of applying inverse model approaches to coastal
waters, a recent study of Moreton and Deception Bays in southeast Queensland,
Australia demonstrated impressive results (Brando and Dekker, 2003). The authors
used hyperspectral Hyperion Imaging Spectrometer data from the NASA Earth
Observing One (EO-1) satellite. A MODTRAN-4 based atmospheric correction for
coastal waters was developed. Validation of the atmospheric correction was confirmed
by establishing good agreement between Hyperion hyperspectral reflectance and
subsurface reflectance measured by a field instrument. The SIOPs of three classes of
optical constituents (chl a, tripton, and CDOM) were measured and validated using
72
Schalles
Reflectance Spectra
The problems with site and temporal specificity in the above approaches using
empirical and semi-analytical algorithms have motivated numerous researchers to
develop analytical model approaches to link IOPs with measured AOPs, including
reflectance spectra (Doerffer and Fischer, 1991; Dekker et al., 1997; Lahet et al., 2001;
Sathyendranath et al., 2001; Pozdnyakov et al., 2002; Brando and Dekker, 2003;
Bowers et al., 2004; Gege and Albert, this volume). If IOPs are well characterized and
field reflectance data are carefully collected, inverse models can accurately retrieve
water constituents (Bukata et al., 1995). However, the inverse approach is highly
sensitive to errors in the measured radiometric variables, and even small errors may
invalidate the inversions (Mobley, 1994).
An obvious problem is finding unique solutions for inversion problems in which
optical water constituents and boundary conditions vary. Potentially, different sets of
boundary and IOP conditions may lead to the same solution. Iterative, Monte Carlo
techniques are typically used to find an optimal solution. Boundary conditions
(atmospheric aerosols, water surface conditions, sun angle) may be very dynamic,
particularly in coastal areas. High accuracy in atmospheric correction is crucial to
success in discriminating water constituents, especially with satellite data (Brando and
Dekker, 2003). The ability to differentiate optically shallow versus optically deep
waters and the contribution of bottom reflectance is also important. Lee et al. (1999)
developed a model which can derive bottom depths and water column properties from
reflectance spectra. Heterogeneity in the IOPs is a further challenge. For example,
differences in the slope of the absorption spectra of CDOM and detrital tripton
(Bricaud et al., 1991; Carder et al., 1989; Dekker, 1993; Bowers et al., 2003),
variability in pigment composition and cell packaging effects, and differences in
volume scattering functions of the different classes of sestonic particles (Dekker et al.,
1997) can all reduce the robustness of the assumed IOPs in coastal settings. Thus, it
becomes necessary to measure and define specific IOPs (SIOPs) for a given site, and
perhaps for a specific time (Brando and Dekker, 2003). Many of the inverse procedures
reduce the optically active constituents to the three primary classes reviewed here (algal
pigment, seston, and CDOM). However, these classes are often, themselves, complex
mixes of constituents. Tidal fluxes, river plumes, and other physical mechanisms,
which create spatial and temporal variability, can complicate the application of
inversion models to large areas measured with satellite sensors (Doerffer et al., 1994).
In many cases, it may not be practical for ship based researchers to synoptically
measure water properties for IOP estimation. In fairness, this set of considerations often
applies to the simpler, band ratio algorithms discussed above.
In spite of the serious constraints of applying inverse model approaches to coastal
waters, a recent study of Moreton and Deception Bays in southeast Queensland,
Australia demonstrated impressive results (Brando and Dekker, 2003). The authors
used hyperspectral Hyperion Imaging Spectrometer data from the NASA Earth
Observing One (EO-1) satellite. A MODTRAN-4 based atmospheric correction for
coastal waters was developed. Validation of the atmospheric correction was confirmed
by establishing good agreement between Hyperion hyperspectral reflectance and
subsurface reflectance measured by a field instrument. The SIOPs of three classes of
optical constituents (chl a, tripton, and CDOM) were measured and validated using
72
Schalles