was simulated from imagery obtained by the Hyperspectral Imager for the Coastal
Ocean (HICO) on the International Space Station. The spectral characteristics of the
simulated Sentinel-2 imagery allowed for accurate measurements of chlorophyll in
optically complex waters in contrast to the limitations encountered with Landsat
8 bands (Fig. 3; see Sect. 2.2.5) which misrepresented SS min as chlorophyll a.
3.1 Empirical and Semi-analytical Approaches
to Lake Water Quality Assessment
The algorithms used to retrieve water quality data for inland waters from satellite
imagery are empirical to semi-analytical. Empirical algorithms statistically model
relationships between measured water quality variables and spectral bands and/or
combinations of spectral bands. Strictly empirical algorithms require no understanding of the physics required to model atmospheric and underwater optical
properties. Thus, they are relatively simple to perform and are well represented in
the literature. This approach is also where many have “oversold” what can be
sensed with remote sensing (i.e., when imagery is used for measurements of variables that have no optical properties, such as phosphorus, DOC, or bacteria [23, 24,
79, 80]).
A better approach involves semi-empirical methods, which use bands that are
selected based on knowledge of how optically active parameters affect reflectance
in various spectral bands. Once such models are identified, they can be applied and
used for routine monitoring, as has been done for water clarity assessments of over
20,000 lakes in Minnesota [32], Wisconsin [81], and Michigan [82, 83]. These
assessments used field data within a few days of the Landsat image acquisition to
calibrate models using the ratio of the Landsat TM1/TM3 bands plus band TM1 as
predictor variables. Matthews [35] recently provided a thorough review of the
literature on empirical and semi-empirical methods using ORS to measure inland
water quality, and that paper should be accessed for further details.
Theoretically, once systemic and atmospheric correction is accurately applied to
imagery allowing for a true water-leaving reflectance product, universal algorithms
could be developed for specific sensors and water quality variables, which would
reduce the need for contemporaneous field data. Unfortunately, accurate atmospheric correction for inland water quality is difficult on a regional basis and
needs further development to be operational [1, 84].
Analytical methods are theoretically derived and use complex approaches such
as radiative transfer or bio-optical modeling, and semi-analytical methods use
analytical techniques that are empirically parameterized with in situ data. Semianalytical methods to estimate water quality variables are thought by some to be the
pathway to global water quality products using ORS [84]. However, many challenges remain with parameterization of the algorithms, and at present there are no
successful validated regional assessments using semi-analytical methods in the
Remote Sensing for Regional Lake Water Quality Assessment: Capabilities and. . .
131
Ocean (HICO) on the International Space Station. The spectral characteristics of the
simulated Sentinel-2 imagery allowed for accurate measurements of chlorophyll in
optically complex waters in contrast to the limitations encountered with Landsat
8 bands (Fig. 3; see Sect. 2.2.5) which misrepresented SS min as chlorophyll a.
3.1 Empirical and Semi-analytical Approaches
to Lake Water Quality Assessment
The algorithms used to retrieve water quality data for inland waters from satellite
imagery are empirical to semi-analytical. Empirical algorithms statistically model
relationships between measured water quality variables and spectral bands and/or
combinations of spectral bands. Strictly empirical algorithms require no understanding of the physics required to model atmospheric and underwater optical
properties. Thus, they are relatively simple to perform and are well represented in
the literature. This approach is also where many have “oversold” what can be
sensed with remote sensing (i.e., when imagery is used for measurements of variables that have no optical properties, such as phosphorus, DOC, or bacteria [23, 24,
79, 80]).
A better approach involves semi-empirical methods, which use bands that are
selected based on knowledge of how optically active parameters affect reflectance
in various spectral bands. Once such models are identified, they can be applied and
used for routine monitoring, as has been done for water clarity assessments of over
20,000 lakes in Minnesota [32], Wisconsin [81], and Michigan [82, 83]. These
assessments used field data within a few days of the Landsat image acquisition to
calibrate models using the ratio of the Landsat TM1/TM3 bands plus band TM1 as
predictor variables. Matthews [35] recently provided a thorough review of the
literature on empirical and semi-empirical methods using ORS to measure inland
water quality, and that paper should be accessed for further details.
Theoretically, once systemic and atmospheric correction is accurately applied to
imagery allowing for a true water-leaving reflectance product, universal algorithms
could be developed for specific sensors and water quality variables, which would
reduce the need for contemporaneous field data. Unfortunately, accurate atmospheric correction for inland water quality is difficult on a regional basis and
needs further development to be operational [1, 84].
Analytical methods are theoretically derived and use complex approaches such
as radiative transfer or bio-optical modeling, and semi-analytical methods use
analytical techniques that are empirically parameterized with in situ data. Semianalytical methods to estimate water quality variables are thought by some to be the
pathway to global water quality products using ORS [84]. However, many challenges remain with parameterization of the algorithms, and at present there are no
successful validated regional assessments using semi-analytical methods in the
Remote Sensing for Regional Lake Water Quality Assessment: Capabilities and. . .
131