Simulated Landsat 8, Sentinel-2, and Sentinel-3 bands calculated from the
hyperspectra yielded r
2
¼ 0.84–0.86 for a 440 . The broader Landsat 8 bands worked
nearly as well as the narrower Sentinel bands and hyperspectral bands, probably
because CDOM lacks specific peaks or troughs in absorbance or reflectance. These
r
2 values generally are considered very good for remote sensing predictive equations. Nonetheless, the average for absolute values of percent difference between
measured and predicted CDOM across the four best predictive models was 31 %.
Although some of the differences can be attributed to sampling variability and
measurement uncertainty, the largest source likely is model error. Such large
uncertainties should serve as a cautionary note to limnologists and remote sensing
scientists.
The effectiveness of predictive equations based on longer wavelengths is counterintuitive given that CDOM absorptivity increases quasi-exponentially with
decreasing wavelength and is increasingly diminished in the green and red regions.
Although the physical basis for the relationships is still uncertain, the higher band
(ALI3, OLI4) centered at 670 nm probably corrects for effects of chlorophyll on
reflectance, and the lower band (ALI2, OLI3) centered at ~560 nm probably
measures the influence of CDOM. It is important to realize that the small absorbance values measured in laboratory spectrophotometers involve much shorter light
paths (1–10 cm) than those of interest in lakes. For example, as noted earlier, a
CDOM level of a 440 ¼ 20 m
À1 implies a Secchi depth of ~1.5 m. Based on
UV-visible absorbance spectra we have measured on similar waters, such a sample
would have an absorbance (A) of ~0.022 at 560 nm in a 1 cm cell. Given that the
Beer-Lambert law applies, the value of A that would apply to a light path of 1.5 m
would be ~3.3, and converting to percent of incident light at 560 nm remaining at a
depth of 1.5 m yields a value of ~3 %. Light reflected back to the air-water interface
from the white surface of the Secchi disk again must travel 1.5 m through the
absorbing medium, and thus much less than 1 % of the incident light at 560 nm
arrives back at the water surface. Clearly, even small absorbance values measured
in the laboratory have large effects on reflectance when the long light paths of lake
water columns are considered.
Brezonik et al. [29] found a CDOM-dependent difference in slopes of reflectance
spectra in the range of ~570–650 nm. For low-CDOM waters, reflectance decreased
with increasing wavelength, but for high-CDOM waters reflectance increased with
increasing wavelength. Even though CDOM absorbance is low in this range, it does
affect reflectance, as the calculation in the preceding paragraph demonstrated.
Other constituents that affect reflectance spectra (notably plant pigments) have
minimal effects in this wavelength region.
CDOM levels are much lower in marine waters than in freshwaters; absorptivity
at 412 nm, a 412 , generally is < 1 m
À1 in coastal waters and < 0.1 m
À1 in the open
ocean. In contrast, a 440 values < ~2 m
À1 in lakes generally are considered negligible, although remote sensing scientists are starting to take interest in measuring
CDOM in low-CDOM lakes [29]. Values in lakes that are considered “humic
colored” commonly are in the range of ~5–20 m
À1 and may range up to 40 m
À1
or even more in highly colored bogs. Concentrations of TSS and chlorophyll also
Remote Sensing for Regional Lake Water Quality Assessment: Capabilities and. . .
125
hyperspectra yielded r
2
¼ 0.84–0.86 for a 440 . The broader Landsat 8 bands worked
nearly as well as the narrower Sentinel bands and hyperspectral bands, probably
because CDOM lacks specific peaks or troughs in absorbance or reflectance. These
r
2 values generally are considered very good for remote sensing predictive equations. Nonetheless, the average for absolute values of percent difference between
measured and predicted CDOM across the four best predictive models was 31 %.
Although some of the differences can be attributed to sampling variability and
measurement uncertainty, the largest source likely is model error. Such large
uncertainties should serve as a cautionary note to limnologists and remote sensing
scientists.
The effectiveness of predictive equations based on longer wavelengths is counterintuitive given that CDOM absorptivity increases quasi-exponentially with
decreasing wavelength and is increasingly diminished in the green and red regions.
Although the physical basis for the relationships is still uncertain, the higher band
(ALI3, OLI4) centered at 670 nm probably corrects for effects of chlorophyll on
reflectance, and the lower band (ALI2, OLI3) centered at ~560 nm probably
measures the influence of CDOM. It is important to realize that the small absorbance values measured in laboratory spectrophotometers involve much shorter light
paths (1–10 cm) than those of interest in lakes. For example, as noted earlier, a
CDOM level of a 440 ¼ 20 m
À1 implies a Secchi depth of ~1.5 m. Based on
UV-visible absorbance spectra we have measured on similar waters, such a sample
would have an absorbance (A) of ~0.022 at 560 nm in a 1 cm cell. Given that the
Beer-Lambert law applies, the value of A that would apply to a light path of 1.5 m
would be ~3.3, and converting to percent of incident light at 560 nm remaining at a
depth of 1.5 m yields a value of ~3 %. Light reflected back to the air-water interface
from the white surface of the Secchi disk again must travel 1.5 m through the
absorbing medium, and thus much less than 1 % of the incident light at 560 nm
arrives back at the water surface. Clearly, even small absorbance values measured
in the laboratory have large effects on reflectance when the long light paths of lake
water columns are considered.
Brezonik et al. [29] found a CDOM-dependent difference in slopes of reflectance
spectra in the range of ~570–650 nm. For low-CDOM waters, reflectance decreased
with increasing wavelength, but for high-CDOM waters reflectance increased with
increasing wavelength. Even though CDOM absorbance is low in this range, it does
affect reflectance, as the calculation in the preceding paragraph demonstrated.
Other constituents that affect reflectance spectra (notably plant pigments) have
minimal effects in this wavelength region.
CDOM levels are much lower in marine waters than in freshwaters; absorptivity
at 412 nm, a 412 , generally is < 1 m
À1 in coastal waters and < 0.1 m
À1 in the open
ocean. In contrast, a 440 values < ~2 m
À1 in lakes generally are considered negligible, although remote sensing scientists are starting to take interest in measuring
CDOM in low-CDOM lakes [29]. Values in lakes that are considered “humic
colored” commonly are in the range of ~5–20 m
À1 and may range up to 40 m
À1
or even more in highly colored bogs. Concentrations of TSS and chlorophyll also
Remote Sensing for Regional Lake Water Quality Assessment: Capabilities and. . .
125