As an example, semi-empirical methods are widely used in productive waters to
assess chl-a concentration. These methods employ band ratios between the secondary chl-a absorption maximum (at around 675 nm) and adjacent spectral bands not
affected by phytoplankton absorption, such as the near-infrared reflectance peak
near 700 nm [26, 27]. In other cases they use a combination of bands in the same red
near-infrared regions [28, 29]. Matthews et al. [15] implemented an algorithm
named ‘Maximum Peak Height’ in order to map the chl-a concentrations from
remotely sensed images in hypertrophic conditions created by cyanobacteria
proliferation.
Spectral inversion procedures are more generic and might be applicable independently of ground measurements and sensor characteristics. In the analytical
approach, the concentrations of water quality parameters are related to the bulk
inherent optical properties (IOPs) via the specific inherent optical properties
(SIOPs). The IOPs of the water column are then related to the water reflectance
and, hence, to the top-of-atmosphere radiance, such as described by the radiative
transfer theory [30, 31]. The analytical method involves inverting all those associations to determine the water quality from remotely sensed data. Approaches used
to invert bio-optical models may include matrix inversion methods [32, 33], neural
networks [34, 35], look-up tables [36, 37], optimisation techniques [38–42] or
classification-based approaches whose end-members are created by forward runs
of the bio-optical model [43]. An example of such approach, using hyperspectral
satellite data, Hyperion, acquired in Case-2 waters for chl-a, tripton and CDOM
retrieval can be found in Brando and Dekker [32] and Giardino et al. [44]. In their
work a linear matrix inversion method was used to invert a bio-optical model,
which was parameterised according to the SIOPs in both study areas, starting from
the atmospherically corrected Hyperion reflectance.
Quantitatively, the relationships developed to assess water quality in lakes
within semi-empirical approaches are often site dependent and can be only applied
to those images from which relationships are derived. Well-calibrated and validated
spectral inversion procedures are instead applicable to every site acquired over the
selected waterbody (presuming constant SIOPs), giving the opportunity to assess
water quality independently from ground measurements. The procedure is also
transferable to other systems, for which the optical characterisation of the
waterbody is known. Nonetheless, they are used less because of the difficulties,
or inaccuracies, in obtaining the parameters and the efforts required in calibrating
the model [45].
Whatever approach is chosen, the assessment of water quality in inland water,
from remote sensing, usually requires the transformation from sensor radiance to
surface reflectance values to remove the atmospheric effect. The atmospheric effect
is a dominant disturbance in remote observation of water quality, and imprecise
corrections may produce large errors in obtaining water reflectance and consequently in retrieving concentrations of water quality parameters [33]. As a minimum, bidirectional effects of scattering and absorption in the atmosphere under
varying aerosol optical depth are usually accounted for. Then to retrieve the water
reflectance, correction for adjacent scattering from land into the light path, and/or
64
C. Giardino et al.
assess chl-a concentration. These methods employ band ratios between the secondary chl-a absorption maximum (at around 675 nm) and adjacent spectral bands not
affected by phytoplankton absorption, such as the near-infrared reflectance peak
near 700 nm [26, 27]. In other cases they use a combination of bands in the same red
near-infrared regions [28, 29]. Matthews et al. [15] implemented an algorithm
named ‘Maximum Peak Height’ in order to map the chl-a concentrations from
remotely sensed images in hypertrophic conditions created by cyanobacteria
proliferation.
Spectral inversion procedures are more generic and might be applicable independently of ground measurements and sensor characteristics. In the analytical
approach, the concentrations of water quality parameters are related to the bulk
inherent optical properties (IOPs) via the specific inherent optical properties
(SIOPs). The IOPs of the water column are then related to the water reflectance
and, hence, to the top-of-atmosphere radiance, such as described by the radiative
transfer theory [30, 31]. The analytical method involves inverting all those associations to determine the water quality from remotely sensed data. Approaches used
to invert bio-optical models may include matrix inversion methods [32, 33], neural
networks [34, 35], look-up tables [36, 37], optimisation techniques [38–42] or
classification-based approaches whose end-members are created by forward runs
of the bio-optical model [43]. An example of such approach, using hyperspectral
satellite data, Hyperion, acquired in Case-2 waters for chl-a, tripton and CDOM
retrieval can be found in Brando and Dekker [32] and Giardino et al. [44]. In their
work a linear matrix inversion method was used to invert a bio-optical model,
which was parameterised according to the SIOPs in both study areas, starting from
the atmospherically corrected Hyperion reflectance.
Quantitatively, the relationships developed to assess water quality in lakes
within semi-empirical approaches are often site dependent and can be only applied
to those images from which relationships are derived. Well-calibrated and validated
spectral inversion procedures are instead applicable to every site acquired over the
selected waterbody (presuming constant SIOPs), giving the opportunity to assess
water quality independently from ground measurements. The procedure is also
transferable to other systems, for which the optical characterisation of the
waterbody is known. Nonetheless, they are used less because of the difficulties,
or inaccuracies, in obtaining the parameters and the efforts required in calibrating
the model [45].
Whatever approach is chosen, the assessment of water quality in inland water,
from remote sensing, usually requires the transformation from sensor radiance to
surface reflectance values to remove the atmospheric effect. The atmospheric effect
is a dominant disturbance in remote observation of water quality, and imprecise
corrections may produce large errors in obtaining water reflectance and consequently in retrieving concentrations of water quality parameters [33]. As a minimum, bidirectional effects of scattering and absorption in the atmosphere under
varying aerosol optical depth are usually accounted for. Then to retrieve the water
reflectance, correction for adjacent scattering from land into the light path, and/or
64
C. Giardino et al.