Factors affecting the measurement of CDOM by remote sensing of optically complex inland waters

Patrick L. Brezonik, Leif Olmanson, Jacques C Finlay, Marvin E. Bauer

Research output: Contribution to journalArticle

88 Citations (Scopus)

Abstract

A combination of new measurements and analysis of historical data from several geographic regions was used to address four issues that affect the reliability and interpretation of colored dissolved organic matter (CDOM) measured by remote sensing of inland waters. First, high variability of CDOM levels in lakes and rivers was found at seasonal and multi-year time scales and at shorter intervals in some rivers and lakes. Coefficients of variation (CVs) of 30%-50% for absorptivity at 440nm (a440) were common in historical and new data sets we examined. CDOM values used to calibrate imagery thus should be measured close to the image acquisition date, preferably within 1-2 months in lakes and a few days in large rivers, unless it can be shown that CDOM levels are temporally stable over longer (or shorter) time periods in a given aquatic system. Second, spectral slopes (S) for CDOM in the visible range vary little over time (even over multi-year periods) within sites. Substantial variation was found between sites, however, and most spectra showed a change in slope near 460nm. Values of S400-460 for waters with moderate to high CDOM levels generally were within a narrow range (~0.014-0.018) and similar to reported S values in the near UV. Values of S400-460 for waters with low CDOM generally were smaller and more variable, as were values for S460-650 for all waters. Overall, the variability of spectral slopes in the visible range should not have a large effect on the reliability of a440 estimates made from remote sensing, which in many models involve reflectance measurements at wavelengths>500nm. Third, although a strong correlation (r2=0.925) was found between CDOM levels and DOC concentrations in 34 surface waters sampled in 2013, the standard error of estimate suggests an uncertainty of~±20% in predicting DOC at a440=5m-1 (a moderate CDOM level). Moreover, CDOM-DOC relationships for unpublished data sets we analyzed and those reported in the literature indicate that both the fraction of DOC that is colored and slopes of regressions between CDOM and DOC are highly variable in space and time. Prediction of DOC concentrations in water bodies from CDOM levels (whether measured in the laboratory or by remote sensing) thus is associated with considerable uncertainty. For the present, this implies that field sampling is required to verify DOC concentrations predicted from remotely sensed CDOM measurements until we have a better understanding of variations in DOC-CDOM relationships. Fourth, shapes of reflectance spectra for CDOM-rich waters varied greatly depending on the concentrations of other constituents (suspended solids and chlorophyll) that affect the optical properties of water. Nonetheless, it is not obvious from our results for several predictive models that different remote sensing algorithms are needed to calculate CDOM levels accurately for waters where CDOM is the only variable affecting reflectance versus waters where other constituents also affect the spectra. The best band or band ratio models for simulated Landsat 8, Sentinel-2 and Sentinel-3 bands from field measured reflectance spectra yielded high r2 values (0.84-0.86) for a440. The broader Landsat 8 bands worked nearly as well for a440 as the narrower Sentinel band sets and hyperspectral bands, probably because CDOM is characterized by a broad exponential increase in absorbance with decreasing wavelength rather than specific peaks or troughs in absorbance or reflectance.

Original languageEnglish (US)
Pages (from-to)199-215
Number of pages17
JournalRemote Sensing of Environment
Volume157
DOIs
StatePublished - Feb 1 2015

Fingerprint

inland waters
dissolved organic matter
Biological materials
remote sensing
Remote sensing
Water
reflectance
Lakes
inland water
Rivers
water
Landsat
absorbance
lakes
rivers
wavelengths
lake
uncertainty
river
wavelength

Keywords

  • CDOM
  • Chlorophyll
  • DOC
  • Hyperspectral imagery
  • Lake superior
  • Lakes
  • Landsat
  • Mississippi river
  • Remote sensing
  • Rivers
  • Sentinel
  • Suspended sediment
  • Turbidity
  • Water quality

Cite this

Factors affecting the measurement of CDOM by remote sensing of optically complex inland waters. / Brezonik, Patrick L.; Olmanson, Leif; Finlay, Jacques C; Bauer, Marvin E.

In: Remote Sensing of Environment, Vol. 157, 01.02.2015, p. 199-215.

Research output: Contribution to journalArticle

@article{242739b43b744d2886f3e535454d5ba6,
title = "Factors affecting the measurement of CDOM by remote sensing of optically complex inland waters",
abstract = "A combination of new measurements and analysis of historical data from several geographic regions was used to address four issues that affect the reliability and interpretation of colored dissolved organic matter (CDOM) measured by remote sensing of inland waters. First, high variability of CDOM levels in lakes and rivers was found at seasonal and multi-year time scales and at shorter intervals in some rivers and lakes. Coefficients of variation (CVs) of 30{\%}-50{\%} for absorptivity at 440nm (a440) were common in historical and new data sets we examined. CDOM values used to calibrate imagery thus should be measured close to the image acquisition date, preferably within 1-2 months in lakes and a few days in large rivers, unless it can be shown that CDOM levels are temporally stable over longer (or shorter) time periods in a given aquatic system. Second, spectral slopes (S) for CDOM in the visible range vary little over time (even over multi-year periods) within sites. Substantial variation was found between sites, however, and most spectra showed a change in slope near 460nm. Values of S400-460 for waters with moderate to high CDOM levels generally were within a narrow range (~0.014-0.018) and similar to reported S values in the near UV. Values of S400-460 for waters with low CDOM generally were smaller and more variable, as were values for S460-650 for all waters. Overall, the variability of spectral slopes in the visible range should not have a large effect on the reliability of a440 estimates made from remote sensing, which in many models involve reflectance measurements at wavelengths>500nm. Third, although a strong correlation (r2=0.925) was found between CDOM levels and DOC concentrations in 34 surface waters sampled in 2013, the standard error of estimate suggests an uncertainty of~±20{\%} in predicting DOC at a440=5m-1 (a moderate CDOM level). Moreover, CDOM-DOC relationships for unpublished data sets we analyzed and those reported in the literature indicate that both the fraction of DOC that is colored and slopes of regressions between CDOM and DOC are highly variable in space and time. Prediction of DOC concentrations in water bodies from CDOM levels (whether measured in the laboratory or by remote sensing) thus is associated with considerable uncertainty. For the present, this implies that field sampling is required to verify DOC concentrations predicted from remotely sensed CDOM measurements until we have a better understanding of variations in DOC-CDOM relationships. Fourth, shapes of reflectance spectra for CDOM-rich waters varied greatly depending on the concentrations of other constituents (suspended solids and chlorophyll) that affect the optical properties of water. Nonetheless, it is not obvious from our results for several predictive models that different remote sensing algorithms are needed to calculate CDOM levels accurately for waters where CDOM is the only variable affecting reflectance versus waters where other constituents also affect the spectra. The best band or band ratio models for simulated Landsat 8, Sentinel-2 and Sentinel-3 bands from field measured reflectance spectra yielded high r2 values (0.84-0.86) for a440. The broader Landsat 8 bands worked nearly as well for a440 as the narrower Sentinel band sets and hyperspectral bands, probably because CDOM is characterized by a broad exponential increase in absorbance with decreasing wavelength rather than specific peaks or troughs in absorbance or reflectance.",
keywords = "CDOM, Chlorophyll, DOC, Hyperspectral imagery, Lake superior, Lakes, Landsat, Mississippi river, Remote sensing, Rivers, Sentinel, Suspended sediment, Turbidity, Water quality",
author = "Brezonik, {Patrick L.} and Leif Olmanson and Finlay, {Jacques C} and Bauer, {Marvin E.}",
year = "2015",
month = "2",
day = "1",
doi = "10.1016/j.rse.2014.04.033",
language = "English (US)",
volume = "157",
pages = "199--215",
journal = "Remote Sensing of Environment",
issn = "0034-4257",
publisher = "Elsevier",

}

TY - JOUR

T1 - Factors affecting the measurement of CDOM by remote sensing of optically complex inland waters

AU - Brezonik, Patrick L.

AU - Olmanson, Leif

AU - Finlay, Jacques C

AU - Bauer, Marvin E.

PY - 2015/2/1

Y1 - 2015/2/1

N2 - A combination of new measurements and analysis of historical data from several geographic regions was used to address four issues that affect the reliability and interpretation of colored dissolved organic matter (CDOM) measured by remote sensing of inland waters. First, high variability of CDOM levels in lakes and rivers was found at seasonal and multi-year time scales and at shorter intervals in some rivers and lakes. Coefficients of variation (CVs) of 30%-50% for absorptivity at 440nm (a440) were common in historical and new data sets we examined. CDOM values used to calibrate imagery thus should be measured close to the image acquisition date, preferably within 1-2 months in lakes and a few days in large rivers, unless it can be shown that CDOM levels are temporally stable over longer (or shorter) time periods in a given aquatic system. Second, spectral slopes (S) for CDOM in the visible range vary little over time (even over multi-year periods) within sites. Substantial variation was found between sites, however, and most spectra showed a change in slope near 460nm. Values of S400-460 for waters with moderate to high CDOM levels generally were within a narrow range (~0.014-0.018) and similar to reported S values in the near UV. Values of S400-460 for waters with low CDOM generally were smaller and more variable, as were values for S460-650 for all waters. Overall, the variability of spectral slopes in the visible range should not have a large effect on the reliability of a440 estimates made from remote sensing, which in many models involve reflectance measurements at wavelengths>500nm. Third, although a strong correlation (r2=0.925) was found between CDOM levels and DOC concentrations in 34 surface waters sampled in 2013, the standard error of estimate suggests an uncertainty of~±20% in predicting DOC at a440=5m-1 (a moderate CDOM level). Moreover, CDOM-DOC relationships for unpublished data sets we analyzed and those reported in the literature indicate that both the fraction of DOC that is colored and slopes of regressions between CDOM and DOC are highly variable in space and time. Prediction of DOC concentrations in water bodies from CDOM levels (whether measured in the laboratory or by remote sensing) thus is associated with considerable uncertainty. For the present, this implies that field sampling is required to verify DOC concentrations predicted from remotely sensed CDOM measurements until we have a better understanding of variations in DOC-CDOM relationships. Fourth, shapes of reflectance spectra for CDOM-rich waters varied greatly depending on the concentrations of other constituents (suspended solids and chlorophyll) that affect the optical properties of water. Nonetheless, it is not obvious from our results for several predictive models that different remote sensing algorithms are needed to calculate CDOM levels accurately for waters where CDOM is the only variable affecting reflectance versus waters where other constituents also affect the spectra. The best band or band ratio models for simulated Landsat 8, Sentinel-2 and Sentinel-3 bands from field measured reflectance spectra yielded high r2 values (0.84-0.86) for a440. The broader Landsat 8 bands worked nearly as well for a440 as the narrower Sentinel band sets and hyperspectral bands, probably because CDOM is characterized by a broad exponential increase in absorbance with decreasing wavelength rather than specific peaks or troughs in absorbance or reflectance.

AB - A combination of new measurements and analysis of historical data from several geographic regions was used to address four issues that affect the reliability and interpretation of colored dissolved organic matter (CDOM) measured by remote sensing of inland waters. First, high variability of CDOM levels in lakes and rivers was found at seasonal and multi-year time scales and at shorter intervals in some rivers and lakes. Coefficients of variation (CVs) of 30%-50% for absorptivity at 440nm (a440) were common in historical and new data sets we examined. CDOM values used to calibrate imagery thus should be measured close to the image acquisition date, preferably within 1-2 months in lakes and a few days in large rivers, unless it can be shown that CDOM levels are temporally stable over longer (or shorter) time periods in a given aquatic system. Second, spectral slopes (S) for CDOM in the visible range vary little over time (even over multi-year periods) within sites. Substantial variation was found between sites, however, and most spectra showed a change in slope near 460nm. Values of S400-460 for waters with moderate to high CDOM levels generally were within a narrow range (~0.014-0.018) and similar to reported S values in the near UV. Values of S400-460 for waters with low CDOM generally were smaller and more variable, as were values for S460-650 for all waters. Overall, the variability of spectral slopes in the visible range should not have a large effect on the reliability of a440 estimates made from remote sensing, which in many models involve reflectance measurements at wavelengths>500nm. Third, although a strong correlation (r2=0.925) was found between CDOM levels and DOC concentrations in 34 surface waters sampled in 2013, the standard error of estimate suggests an uncertainty of~±20% in predicting DOC at a440=5m-1 (a moderate CDOM level). Moreover, CDOM-DOC relationships for unpublished data sets we analyzed and those reported in the literature indicate that both the fraction of DOC that is colored and slopes of regressions between CDOM and DOC are highly variable in space and time. Prediction of DOC concentrations in water bodies from CDOM levels (whether measured in the laboratory or by remote sensing) thus is associated with considerable uncertainty. For the present, this implies that field sampling is required to verify DOC concentrations predicted from remotely sensed CDOM measurements until we have a better understanding of variations in DOC-CDOM relationships. Fourth, shapes of reflectance spectra for CDOM-rich waters varied greatly depending on the concentrations of other constituents (suspended solids and chlorophyll) that affect the optical properties of water. Nonetheless, it is not obvious from our results for several predictive models that different remote sensing algorithms are needed to calculate CDOM levels accurately for waters where CDOM is the only variable affecting reflectance versus waters where other constituents also affect the spectra. The best band or band ratio models for simulated Landsat 8, Sentinel-2 and Sentinel-3 bands from field measured reflectance spectra yielded high r2 values (0.84-0.86) for a440. The broader Landsat 8 bands worked nearly as well for a440 as the narrower Sentinel band sets and hyperspectral bands, probably because CDOM is characterized by a broad exponential increase in absorbance with decreasing wavelength rather than specific peaks or troughs in absorbance or reflectance.

KW - CDOM

KW - Chlorophyll

KW - DOC

KW - Hyperspectral imagery

KW - Lake superior

KW - Lakes

KW - Landsat

KW - Mississippi river

KW - Remote sensing

KW - Rivers

KW - Sentinel

KW - Suspended sediment

KW - Turbidity

KW - Water quality

UR - http://www.scopus.com/inward/record.url?scp=85027953631&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85027953631&partnerID=8YFLogxK

U2 - 10.1016/j.rse.2014.04.033

DO - 10.1016/j.rse.2014.04.033

M3 - Article

VL - 157

SP - 199

EP - 215

JO - Remote Sensing of Environment

JF - Remote Sensing of Environment

SN - 0034-4257

ER -