University of Nebraska - Lincoln
DigitalCommons@University of Nebraska - Lincoln
Dissertations & Theses in Natural Resources
Natural Resources, School of
5-3-2012
Near Infrared-red Models for the Remote
Estimation of Chlorophyll-α Concentration in
Optically Complex Turbid Productive Waters:
From in Situ Measurements to Aerial Imagery
Daniela Gurlin
University of Nebraska-Lincoln, [email protected]
Follow this and additional works at: http://digitalcommons.unl.edu/natresdiss
Part of the Natural Resources and Conservation Commons
Gurlin, Daniela, "Near Infrared-red Models for the Remote Estimation of Chlorophyll-α Concentration in Optically Complex Turbid
Productive Waters: From in Situ Measurements to Aerial Imagery" (2012). Dissertations & Theses in Natural Resources. Paper 46.
http://digitalcommons.unl.edu/natresdiss/46
This Article is brought to you for free and open access by the Natural Resources, School of at DigitalCommons@University of Nebraska - Lincoln. It
has been accepted for inclusion in Dissertations & Theses in Natural Resources by an authorized administrator of DigitalCommons@University of
Nebraska - Lincoln.
NEAR INFRARED-RED MODELS
FOR THE REMOTE ESTIMATION OF CHLOROPHYLL-a CONCENTRATION
IN OPTICALLY COMPLEX TURBID PRODUCTIVE WATERS:
FROM IN SITU MEASUREMENTS TO AERIAL IMAGERY
by
Daniela Gurlin
A DISSERTATION
Presented to the Faculty of
The Graduate College at the University of Nebraska
In Partial Fulfillment of Requirements
For the Degree of Doctor of Philosophy
Major: Natural Resource Sciences
Under the Supervision of Professor Anatoly A. Gitelson
Lincoln, Nebraska
May, 2012
NEAR INFRARED-RED MODELS
FOR THE REMOTE ESTIMATION OF CHLOROPHYLL-a CONCENTRATION
IN OPTICALLY COMPLEX TURBID PRODUCTIVE WATERS:
FROM IN SITU MEASUREMENTS TO AERIAL IMAGERY
Daniela Gurlin, Ph.D.
University of Nebraska, 2012
Advisor: Anatoly A. Gitelson
Today the water quality of many inland and coastal waters is compromised by
cultural eutrophication in consequence of increased human agricultural and industrial
activities and remote sensing is widely applied to monitor the trophic state of these
waters. This study explores near infrared-red models for the remote estimation of
chlorophyll-a concentration in turbid productive waters and compares several near
infrared-red models developed within the last 35 years.
Three of these near infrared-red models were calibrated for a dataset with
chlorophyll-a concentrations from 2.3 to 81.2 mg m-3 and validated for independent and
statistically significantly different datasets with chlorophyll-a concentrations from 4.0 to
95.5 mg m-3 and 4.0 to 24.2 mg m-3 for the spectral bands of the MEdium Resolution
Imaging Spectrometer (MERIS) and Moderate-resolution Imaging Spectroradiometer
(MODIS). The developed MERIS two-band algorithm estimated chlorophyll-a
concentrations from 4.0 to 24.2 mg m-3, which are typical for many inland and coastal
waters, very accurately with a mean absolute error 1.2 mg m-3. These results indicate a
high potential of the simple MERIS two-band algorithm for the reliable estimation of
chlorophyll-a concentration without any reduction in accuracy compared to more
complex algorithms, even though more research seems required to analyze the sensitivity
of this algorithm to differences in the chlorophyll-a specific absorption coefficient of
phytoplankton.
Three near infrared-red models were calibrated and validated for a smaller dataset
of atmospherically corrected multi-temporal aerial imagery collected by the hyperspectral
airborne imaging spectrometer for applications (AisaEAGLE). The developed algorithms
successfully captured the spatial and temporal variability of the chlorophyll-a
concentrations and estimated chlorophyll-a concentrations from 2.3 to 81.2 mg m-3 with
mean absolute errors from 4.4 mg m-3 for the AISA two band algorithm to 5.2 mg m-3 for
the AISA three band algorithm.
iv
© 2012 Daniela Gurlin
v
This dissertation is dedicated to
my husband James
and my beautiful Himalayan mix Jake.
vi
Acknowledgements
This research was made possible through the collaborative effort of many
individuals. I would like to thank my advisor and chairman of my supervisory committee
at the University of Nebraska, Prof. Anatoly A. Gitelson, for his support throughout my
studies. I admire his enthusiasm for research and his patience when we encountered
technical difficulties in this research. My supervisory committee member Prof. Donald C.
Rundquist provided a positive work environment at the Center for Advanced Land
Management Information Technology (CALMIT) and continuous support for this
research. I acknowledge the contributions from my supervisory committee members Prof.
Kyle Hoagland and Prof. Merlin Lawson. Prof. Kyle Hoagland provided phycological
expertise and Prof. Merlin Lawson helped me to see the potential applications of my
research from a different perspective.
Bryan Leavitt from CALMIT provided technical support in the field and many
photographs from the field data collection. His technical expertise was invaluable for the
field data collection and he fixed technical problems in the field within a very short time.
I am indebted to Tadd Barrow and Jennifer Swanson from the Aquatic Ecology
and Limnology Laboratory at the University of Nebraska for their assistance in the field
and laboratory. I would like to thank Prof. Lin Li and his students from Indiana
University-Purdue University in Indianapolis for the analysis of the phycocyanin
concentrations.
vii
Dr. Wesley Moses, currently a research associate with The National Research
Council at the Naval Research Laboratory in Washington, DC, was in the second year of
his PhD program when I came to Lincoln and we collected field data for two years at the
Fremont Lakes in a collaborative effort. He processed the in situ reflectance data and
provided the atmospherically corrected hyperspectral aircraft data. Rick Perk from
CALMIT provided the pre-processed aircraft data.
Yi Peng and Tony Nguy-Robertson provided support in the collection of field
data. Yi was involved in some of the laboratory work and filtered many of the water
samples we collected at the Fremont Lakes.
The students at CALMIT created a friendly work atmosphere and we discussed
anything from the correct forms to file throughout the program of studies to fashion
advice for the dissertation defense.
I would like to thank my parents Juergen and Angelika Gurlin for their love and
support and my husband James Andrew Howell for his love, support, and patience. He
never complained when I worked late at night and we lived from frozen pizza for weeks.
Grant Information
This research was supported by US Environmental Protection Agency grant
R-828634501, and the National Aeronautics and Space Administration (NASA) Land
Cover Land Use Change (LCLUC) program grants NNG06GA92G and NNG06GG17G.
viii
Table of Contents
Chapter 1. Introduction ....................................................................................................... 1
1.1
Productivity of inland and coastal waters ............................................................... 2
1.2
Remote estimation of chl-a concentrations in inland and coastal waters ............... 3
1.3
Remote estimation of phycocyanin concentrations in inland waters .................... 10
1.4
Objective ............................................................................................................... 13
1.5
Structure ................................................................................................................ 14
1.6
Publications ........................................................................................................... 16
Chapter 2. Datasets and Methods...................................................................................... 17
2.1
Study area.............................................................................................................. 17
2.2
Field measurements .............................................................................................. 20
2.2.1
Bio-physical and bio-optical water quality parameters .............................................. 20
2.2.2
Reflectance ................................................................................................................. 23
2.3
Laboratory measurements ..................................................................................... 25
2.4
Calibration and validation of the NIR-red models for the remote estimation of
chl-a concentrations .............................................................................................. 33
2.5
Application of NIR-red models for the remote estimation of chl-a concentrations
from multi-temporal AISA imagery ..................................................................... 35
Chapter 3. Bio-physical and Bio-optical Properties of the Fremont Lakes and Branched
Oak Lake ......................................................................................................... 37
3.1
Bio-physical properties of the Fremont Lakes ...................................................... 37
3.2
Inherent bio-optical properties of the Fremont Lakes........................................... 43
3.3
Apparent bio-optical properties of the Fremont Lakes ......................................... 52
3.4
Bio-physical properties of Branched Oak Lake .................................................... 55
3.5
Inherent and apparent bio-optical properties of Branched Oak Lake ................... 57
Chapter 4. Return to a Simple Two-band NIR-red Model for the Remote Estimation of
Chl-a in Turbid Productive Waters? ............................................................... 62
4.1
Importance of the red and NIR spectral regions for the remote estimation of chl-a
concentrations in inland and coastal waters .......................................................... 62
4.2
Empirical and semi-empirical NIR-red models for the remote estimation of chl-a
concentrations in inland and coastal waters .......................................................... 67
ix
4.3
Calibration of the NIR-red models for the remote estimation of low-to-moderate
chl-a concentrations .............................................................................................. 74
4.4
Validation of the NIR-red models for the estimation of low-to-moderate chl-a
concentrations ....................................................................................................... 77
4.4.1
MERIS three-band model and MERIS two-band model ............................................ 77
4.4.2
Comparison of the enhanced three-band index with the MERIS three-band model and
MERIS two-band model............................................................................................. 81
4.4.3
MODIS two-band model ............................................................................................ 83
4.4.4
Comparison of Gons’ NIR-red model and the MERIS three-band and two-band
models ........................................................................................................................ 84
4.4.5
Advanced MERIS three-band model and Advanced MERIS two-band model ......... 88
4.5
Comparison of the performances of the NIR-red algorithms ............................... 91
4.6
Validity of the model calibrations in turbid highly productive inland waters ...... 93
4.7
Estimation of phycocyanin with Simis’ red-NIR algorithm ................................. 95
Chapter 5. Remote Estimation of Chl-a in Turbid Productive Waters from Multi-temporal
AISA Imagery ............................................................................................... 103
5.1
Bio-physical and bio-optical water quality parameters of the Fremont Lakes for
the AISA data collection dates ............................................................................ 103
5.2
Comparison of AISA reflectance spectra and in situ reflectance spectra ........... 105
5.3
Calibration of the NIR-red models for the estimation of chl-a concentrations from
AISA imagery ..................................................................................................... 108
5.4
Validation of the NIR-red models for the estimation of chl-a concentrations for
AISA imagery ..................................................................................................... 111
5.5
Comparison of Gon’s model and the MERIS two-band model for the estimation
of chl-a concentrations for AISA imagery.......................................................... 113
Chapter 6. Conclusions and Future Work ....................................................................... 118
References .................................................................................................................... 121
x
List of Figures
Figure 2.1 Map of the study area at the Fremont Lakes in eastern Nebraska, USA. The
positions of stations are indicated with the red markers. ................................ 17
Figure 2.2 Fremont Lake 07 in summer 2008, photo by B. Leavitt, CALMIT, UNL. .... 18
Figure 2.3 The turbidity provided a first estimate of the filtration volumes (Vf) for the
TSS measurements (Vf shown for lakes 01, 02, 03, 07, and 16 in 2008 and
2009). .............................................................................................................. 22
Figure 2.4 Standard curve from 08/26/2008 for chl-a concentrations from 0 to 200 mg m-3.
...................................................................................................................... 27
Figure 2.5 Variability of the fluorescence readings of the “high” and “low” solid chl-a
standards after the fluorometer calibration on 08/26/2008. ............................ 27
Figure 2.6 Variability of the fluorescence readings of the “high” and “low” solid chl-a
standards after the fluorometer calibration on 08/26/2008 shown in more
detail. ............................................................................................................... 28
Figure 2.7 The distinctive colors of the dried filters indicate substantial differences in the
composition of the suspended solids for these five lakes on 10/09/2008. ...... 29
Figure 2.8 Measurements of the optical density of the particles retained on the filters for
FSL07 on 06/15/2009. The absorption by pigments was successfully removed
through the sodium hypochlorite solution. ..................................................... 31
Figure 2.9 Measurements of the optical densities of the particles retained on the filters
for a station at Branched Oak Lake on 08/18/2011 (a), after correction (b),
and difference in the optical density for NAP (c). .......................................... 31
Figure 3.1 Comparison of the spectrophotometrically and fluorometrically measured chl-a
concentrations for the datasets collected at the Fremont Lakes in 2008 and
2009................................................................................................................. 37
Figure 3.2 Frequency distributions of the chl-a concentrations measured in 2008 (a) and
2009 (b). .......................................................................................................... 38
Figure 3.3 Relationship of TSS concentration and Secchi disk depth for the Fremont
Lakes in 2008 and 2009. ................................................................................. 40
Figure 3.4 Relationship of chl-a concentration and ISS/TSS. The four stations with a
high contribution of ISS to TSS in the presence of high chl-a concentrations
are shown in red. ............................................................................................. 41
Figure 3.5 Seasonal variations in the concentrations of chl-a, ISS, and OSS in Fremont
Lake 02 in summer and fall 2008. .................................................................. 42
xi
Figure 3.6 Representative spectra of the absorption coefficients of total particulates (a),
non-algal particles (b), phytoplankton (c), and CDOM (d) for 21 stations with
chl-a concentrations from 2.3 to 132.4 mg m-3. .............................................. 43
Figure 3.7 Relationship of aφ(675) and chl-a concentration for the Fremont Lakes in
2008 and 2009. ................................................................................................ 44
Figure 3.8 Representative spectra of the chl-a specific absorption coefficient of
phytoplankton (a) and normalized chl-a specific absorption coefficient of
phytoplankton (b) for the same stations shown in Fig. 3.6. The normalization
of a*φ(λ) to a*φ(440) enhances the spectral absorption features of different
types of chlorophylls, carotenoids, the cyanobacterial pigment phycocyanin
(PC), and, for one lake in fall 2008, phycoerythrin (PE). ............................... 46
Figure 3.9 Chl-a specific absorption coefficients of phytoplankton at 400 and 675 nm,
a*φ(440) and a*φ(675), shown versus chl-a concentration for the Fremont
Lakes 2008 and 2009 datasets......................................................................... 47
Figure 3.10 Ternary plots of the absorption budgets at 440 and 675 nm for 2008 (a, b)
and 2009 (c, d). The relative contribution of CDOM, phytoplankton, and NAP
to absorption was calculated through normalization of the individual
absorption components by x(x + y + z)-1, where x is the component of interest,
and y and z designate the second and third component. The distances from the
apices of the triangle indicate the relative contribution of the absorption
components to absorption. .............................................................................. 48
Figure 3.11 Relationship of aCDOM(440) and SCDOM for the Fremont Lakes in 2008 and
2009................................................................................................................. 49
Figure 3.12 Frequency distributions of the slope of CDOM, SCDOM, measured in 2008
and 2009. The standard normal distribution curve for the European coastal
waters studied by Babin et al. (2003) for an average SCDOM of 0.0176 nm-1 and
a standard deviation of 0.002 nm-1 is shown for comparison. ........................ 50
Figure 3.13 Spectra of the backscattering coefficient, bb(λ), collected in 2009. Only the
spectra without any saturation of the ECO Triplet sensor are shown. The
sensor was typically saturated in the NIR spectral region at a turbidity of 8 to
10 NTU and the stations affected by the saturation of the sensor are expected
to have a higher bb(λ). ..................................................................................... 51
Figure 3.14 Relationship of the particulate backscattering coefficient, bbp(660), with ISS
and OSS. ......................................................................................................... 52
Figure 3.15 Remote sensing reflectance, ρrs(λ), spectra for the waters sampled in 2008
(a) and 2009 (b). The coefficient of variation of the reflectance is shown in
grey. ................................................................................................................ 53
xii
Figure 3.16 Inverse reflectance spectrum for a station with a chl-a concentration of 119.7
mg m-3. The positions of the maximal absorption by phytoplankton pigments,
NAP, and CDOM, the position of the minimal absorption by phytoplankton
pigments (Min 1), and the position of the minimal combined absorption by
chl-a, NAP, and water (Min 2) are indicated by arrows. ................................ 54
Figure 3.17 Frequency distributions of the chl-a concentrations measured in 2008 and
2009 in the Fremont Lakes and in 2011 in Branched Oak Lake. The line
represents the standard normal distribution curve for Branched Oak Lake.... 55
Figure 3.18 Comparison of the relationship of chl-a concentration and ISS/TSS for the
Fremont Lakes and Branched Oak Lake. ........................................................ 56
Figure 3.19 Representative spectra of the absorption coefficients of total particulates (a),
non-algal particles (b), phytoplankton (c), and CDOM (d) for 21 stations with
chl-a concentrations from 83.6 to 159.7 mg m-3. ............................................ 57
Figure 3.20 Representative spectra of the chl-a specific absorption coefficient of
phytoplankton for the same stations shown in Fig. 3.20................................. 58
Figure 3.21 Relationship of the chl-a specific absorption coefficients of phytoplankton at
400 and 675 nm, a*φ(440) and a*φ(675), and chl-a concentration. ................. 59
Figure 3.22 Ternary plots of the absorption budgets at 440 and 675 nm for the Fremont
Lakes and Branched Oak Lake. ...................................................................... 60
Figure 3.23 Remote sensing reflectance, ρrs(λ), spectra for the waters sampled in 2011.
The coefficient of variation of the reflectance is shown in grey..................... 61
Figure 4.1 Comparison of the absorption coefficients of phytoplankton, aφ(λ), non-algal
particles, aNAP(λ), CDOM, aCDOM(λ), and water, aw(λ), for two stations with
chl-a concentrations of 4.6 mg m-3 (a) and 58.1 mg m-3 (b). The chl-a
absorption peak at 440 nm for the station with a chl-a concentration of 4.6 mg
m-3 is completely concealed by the absorption by NAP and CDOM (aw(λ)
taken from Mueller, 2003). ............................................................................. 63
Figure 4.2. Application of the MERIS OC4E algorithm for the estimation of chl-a in
turbid inland waters......................................................................................... 64
Figure 4.3 Spectra of the combined absorption coefficients of phytoplankton and water
for chl-a concentrations from 2.3 to 132.4 mg m-3 (a), and the position of the
minimal combined absorption by chl-a and water in the red region (b). The
blue line represents the absorption coefficient of water, aw(λ), (aw(λ) taken
from Mueller (2003). ...................................................................................... 66
Figure 4.4 Relationship of chl-a concentration and reflectance peak position for the
Fremont Lakes 2008 and 2009 datasets. ......................................................... 67
xiii
Figure 4.5 Relationship of chl-a concentration and fluorescence line height for the
Fremont Lakes 2008 dataset. .......................................................................... 68
Figure 4.6 Comparison of the fluorescence line height, FLH (a, c), and reflectance line
height, RLH (b, d), for a lake with a chl-a concentration of 18.0 mg m-3. ..... 69
Figure 4.7 Relationships of chl-a concentration with reflectance line height (a) and
normalized reflectance line height (b) for the Fremont Lakes 2008 dataset. . 69
Figure 4.8 Relationship of ρ(705)/ρ(670) and chl-a concentration. The station marked in
red was excluded from the quadratic regression. ............................................ 70
Figure 4.9 Identification of the optimal spectral regions for the estimation of chl-a with
the three-band model (Eq. 1.5). The identification of λ1, λ2, and λ3 for the
three-band model involved the reduction of the standard error of the estimate,
STE, in an iterative process. The vertical lines represent the spectral regions
with the smallest possible STE. ...................................................................... 72
Figure 4.10 Identification of the optimal spectral regions for the estimation of chl-a with
the two-band model (Eq. 1.7). ........................................................................ 73
Figure 4.11 Calibration of the Eq. 4.1 MERIS three-band model (a), Eq. 4.2 MERIS twoband model (b), and the Eq. 4.3 MODIS two-band model (c), for the
estimation of chl-a concentrations up to 85 mg m-3. ...................................... 76
Figure 4.12 Application of the MERIS three-band model, calibrated from the Fremont
Lakes 2008 dataset, for the estimation of chl-a concentrations up to 100 mg m-3
for the Fremont Lakes 2009 dataset. ............................................................... 78
Figure 4.13 Application of the MERIS two-band model, calibrated from the Fremont
Lakes 2008 dataset, for the estimation of chl-a concentrations up to 100 mg m-3
for the Fremont Lakes 2009 dataset. ............................................................... 78
Figure 4.14 Application of the MERIS three-band model, calibrated from the Fremont
Lakes 2008 dataset, for the estimation of chl-a concentrations up to 25 mg m-3
for the Fremont Lakes 2009 dataset. ............................................................... 79
Figure 4.15 Application of the MERIS two-band model, calibrated from the Fremont
Lakes 2008 dataset, for the estimation of chl-a concentrations up to 25 mg m-3
for the Fremont Lakes 2009 dataset. ............................................................... 79
Figure 4.16 Relationship of TSS concentration and remote sensing reflectance, Rrs(λ), at
665, 709, and 754 nm for stations with chl-a concentrations from 4.0 to 24.2
mg m-3 in 2009. ............................................................................................... 80
xiv
Figure 4.17 Application of the enhanced three-band index, calibrated for Lake
Kasumigure, Japan (Yang et al., 2010), for the estimation of chl-a
concentrations up to 100 mg m-3 (a) and up to 25 mg m-3 (b) for the Fremont
Lakes 2009 dataset. ......................................................................................... 82
Figure 4.18 Application of the MODIS two-band model, calibrated from the Fremont
Lakes 2008 dataset, for the estimation of chl-a concentrations up to 100 mg m-3
(a) and up to 25 mg m-3 (b) for the Fremont Lakes 2009 dataset. .................. 83
Figure 4.19 Comparison of Gons’ original algorithm (a), (Eq. 4.8), and re-parameterized
algorithm (b), (Eq. 4.9), for the estimation of chl-a concentrations up to 100
mg m-3 for the Fremont Lakes 2009 dataset. ................................................... 86
Figure 4.20 Comparison of ρ(665)-1 × ρ(709) and the chl-a concentrations estimated
through Gons’ re-parameterized algorithm (Eq. 4.9)...................................... 87
Figure 4.21 Application of the advanced MERIS three-band model (Gilerson et al.,
2010) for the estimation of chl-a concentrations up to 100 mg m-3 for the
Fremont Lakes 2009 dataset. .......................................................................... 90
Figure 4.22 Relationship of chl-a concentrations and MERIS three-band model values for
the Fremont Lakes 2008 dataset, used for the algorithm development, and the
Branched Oak Lake 2011 dataset. The linear regression line is representative
of the combined dataset. ................................................................................. 93
Figure 4.23 Relationship of chl-a concentrations and MERIS two-band model values for
the Fremont Lakes 2008 and Branched Oak Lake 2011 datasets. The linear
regression line is representative of the combined dataset. .............................. 94
Figure 4.24 Relationship of chl-a concentrations and MODIS two-band model values for
the Fremont Lakes 2008 and Branched Oak Lake 2011 datasets. The offsets
and intercepts of the regression lines for the Fremont Lakes 2008 dataset and
the combined dataset are practically identical. ............................................... 94
Figure 4.25 Identification of γ in Eq. 4.14 through a linear least squares fit of the
measured and estimated (Eq. 4.14 with γ = 1) aφ(665). The intercept was set to
zero. ................................................................................................................. 97
Figure 4.26 Application of a variation of Gons’ algorithm (Eq. 4.14) for the estimation of
chl-a concentrations up to 100 mg m-3 for the Fremont Lakes 2009 dataset.
The algorithm was re-parameterized with γ = 0.9986 and the median
analytically measured a*chl-a(665) value of 0.0112 for the Fremont Lakes 2008
dataset. ............................................................................................................ 98
Figure 4.27 Application of Simis’ algorithm (Eqs. 4.15 and 4.16) for the estimation of
phycocyanin concentrations for the Fremont Lakes 2009 dataset. The two
outliers marked in red were excluded from the calculation of the statistics. .. 99
xv
Figure 4.28 Identification of δ in Eq. 4.15 through a linear least squares fit of the
measured and estimated (Eq. 4.15 with δ = 1) aφ(620). The intercept was set to
zero. ............................................................................................................... 100
Figure 4.29 Relationship of PC/Chl-a and analytically measured phycocyanin specific
absorption coefficient of phytoplankton, a*PC(620). The relationship is
different from the relationship expected for Cyanobacteria dominated waters
for which a relatively constant a*PC(620) is typical. The two previously
identified outliers have PC/Chl-a ratios of 0.0035 and 1.5568 and a a*PC(620)
of 0.0070 and 0.9469 and fit the relationship shown. ................................... 101
Figure 4.30 Application of Simis’ re-parameterized algorithm (Simis et al., 2005) for the
estimation of phycocyanin concentrations for the Fremont Lakes 2009 dataset.
The two outliers were excluded. ................................................................... 102
Figure 5.1 Chl-a specific absorption coefficient of phytoplankton at 675 nm, a*φ(675),
shown versus chl-a concentration. ................................................................ 105
Figure 5.2 Remote sensing reflectance, ρrs(λ), spectra derived from the AISA
measurements. The coefficient of variation of the reflectance is shown in grey.
.................................................................................................................... 106
Figure 5.3 Comparison of the QUAC corrected AISA spectra (solid lines) and the in situ
measured Ocean Optics spectra (dashed lines) for three stations with chl-a
concentrations of 7.6 mg m-3 (a), 25.2 mg m-3 (b), and 81.2 mg m-3 (c). ..... 107
Figure 5.4 Calibration of the Eq. 5.1 AISA three-band model (a), and Eq. 5.2 AISA twoband model (b) for the estimation of chl-a concentrations from the QUAC
corrected AISA images. ................................................................................ 110
Figure 5.5 Application of the AISA three-band model, calibrated from the QUAC
corrected AISA images taken on 07/14/2008 and 10/25/2008, for the
estimation of chl-a concentrations for the images taken on 07/02/2008,
09/26/2008, and 11/19/2008. ........................................................................ 112
Figure 5.6 Application of the AISA two-band model, calibrated from the QUAC
corrected AISA images taken on 07/14/2008 and 10/25/2008, for the
estimation of chl-a concentrations for the images taken on 07/02/2008,
09/26/2008, and 11/19/2008. ........................................................................ 112
Figure 5.7 Application of the Gons’ model, calibrated from the QUAC corrected AISA
images taken on 07/14/2008 and 10/25/2008, for the estimation of chl-a
concentrations for the images taken on 07/02/2008, 09/26/2008, and
11/19/2008. ................................................................................................... 114
Figure 5.8 Map of the chl-a distribution in the Fremont Lakes for 07/02/2008 produced
from the AISA two-band algorithm (Eq. 5.4). .............................................. 116
xvi
Figure 5.9 Map of the chl-a distribution in the Fremont Lakes for 09/26/2008 produced
from the AISA two-band algorithm (Eq. 5.4). .............................................. 116
Figure 5.10 Map of the chl-a distribution in the Fremont Lakes for 11/19/2008 produced
from the AISA two-band algorithm (Eq. 5.4). .............................................. 117
List of Tables
Table 2.1 Filtration volumes for the laboratory analysis of the chl-a, -b, and –c
concentrations, phycocyanin concentrations, absorption coefficients, and
suspended solids for 06/15/2009. .................................................................... 22
Table 2.2 Spectral bands of the MERIS and MODIS satellite sensors used for the
development of NIR-red algorithms for the remote estimation of chl-a
concentrations. ................................................................................................ 34
Table 3.1 Descriptive statistics of the water quality parameters measured in 2008. ....... 39
Table 3.2 Descriptive statistics of the water quality parameters measured in 2009. ....... 39
Table 3.3 Descriptive statistics of the water quality parameters measured in 2011. ....... 56
Table 4.1 Performance of the NIR-red algorithms for the estimation of chl-a
concentrations up to 100 mg m-3. .................................................................... 92
Table 4.2 Performance of the NIR-red algorithms for the estimation of chl-a
concentrations up to 25 mg m-3. ...................................................................... 92
Table 4.3 Descriptive statistics of the water quality parameters ...................................... 96
Table 5.1 Descriptive statistics of the water quality parameters measured at the times of
AISA data collection in 2008. ....................................................................... 104
Table 5.2 Performance of the NIR-red algorithms for the estimation of chl-a
concentrations for the QUAC corrected AISA imagery ............................... 115
1
Chapter 1. Introduction
Inland and coastal waters represent complex ecosystems and provide a multitude
of ecosystem services (Costanza et al., 1997) to human populations. Due to the essential
nature of these services for human welfare, the sustainable management of these
ecosystems represents many environmental and political challenges, and the importance
of the protection of water resources, water quality, and aquatic ecosystems was addressed
in Agenda 21 (United Nations Conference on Environment and Development (UNCED)
1992) and the World Water Development Reports (World Water Assessment Programme
2003, 2006, and 2009). With many aquatic ecosystems threatened by the destruction of
catchment areas, deforestation, inadequately treated domestic sewage and industrial
wastewaters, and the extensive agricultural application of fertilizers, pesticides, and
herbicides, the protection of these ecosystems is considered essential to maintain
environmental and human health (UNCED, 1992). The development of new strategies for
the sustainable management of inland and coastal waters might prevent costly measures
for the rehabilitation, treatment, and development of new water supplies in the future
(UNCED, 1992). Remote sensing is widely applied to monitor the trophic state of inland
and coastal waters (Dekker, 1991; Mittenzwey et al., 1991; Dekker, 1993; Baban, 1993
and 1996, Kallio et al., 2001; Strömbeck and Pierson, 2001; Koponen et al., 2002;
Thiemann and Kaufmann, 2002; Koponen et al., 2007; Gitelson et al., 2008; Gons et al.,
2008) and provides important information for the development of new strategies for the
sustainable management of these ecosystems (Strand et al., 2007).
2
1.1
Productivity of inland and coastal waters
Cultural eutrophication from the input of nitrogen and phosphorous from
agricultural runoff, domestic sewage, industrial wastewaters, and the atmosphere is one
of the most prevalent water quality problems worldwide (Revenga and Kura, 2003;
World Water Assessment Programme, 2009). It is reflected in an increase in productivity
(Goldman, 1968; Vollenweider, 1968; Thomas, 1969; Schindler, 1978; Goldman, 1988;
Carpenter et al., 1996; Schindler; 2006) and changes in the species composition of the
phytoplankton (Reynolds, 1987; Reynolds, 1998) with possible blooms of cyanobacteria
(Carpenter et al., 1998) in freshwater ecosystems. The increased productivity of inland
and coastal waters may interfere with the use of water for fisheries, agriculture, industry,
recreation, and human consumption (Carpenter et al., 1998) and toxic algae blooms in
coastal waters may have catastrophic economic impacts on aquaculture and shellfisheries
(Shumway, 1990). The decomposition of aquatic plants and phytoplankton leads to
oxygen depletion in many inland and coastal waters in summer and fall and may cause
fish kills (Carpenter et al., 1998).
The concentration of chlorophyll-a (chl-a), a pigment found in every
phytoplankton species, is a measure of the productivity of waters and one of the standard
water quality parameters for the evaluation of the trophic state of inland and coastal waters
(Carlson, 1977; Lillesand et al., 1983; Baban, 1996; Thiemann and Kaufmann, 2002). The
estimation of chl-a concentrations from remotely sensed data requires the development of
algorithms with a maximal sensitivity to the concentration of chl-a and minimal
sensitivity to the concentrations of the rest of the constituents present in the water.
3
1.2
Remote estimation of chl-a concentrations in inland and
coastal waters
Many researchers initially concentrated on the interpretation of remotely sensed
ocean color and the development of algorithms for oceanic Case I waters (Clark, 1981;
Smith and Wilson, 1981; Gordon et al., 1983; Gordon and Morel, 1983; Gordon et al.,
1988). In typical Case I waters, the reflectance is influenced by absorption and scattering
by phytoplankton particles, the associated debris from grazing by zooplankton and
natural decay, and colored dissolved organic matter (CDOM) released from the debris
(Morel and Prieur, 1977; Bricaud et al., 1981; Gordon and Morel, 1983). Case I waters
comprise the open ocean and are found in coastal areas, without a continental shelf and
input of terrigenous particles, and in some upwelling regions (Gordon and Morel, 1983).
Contrary to Case I waters, the reflectance in Case II waters is controlled by inorganic
particles and the contribution of phytoplankton pigments to total absorption in ideal Case
II waters is minimal (Morel and Prieur, 1977). Reflectance of Case II waters is influenced
by re-suspended sediments, terrigenous particles and CDOM, and particulate and
dissolved materials of anthropogenic origin (Gordon and Morel, 1983).
Inland and coastal waters are optically complex and may combine the optical
properties of Case I and Case II waters. Reflectance of these waters is controlled by the
combined effects of absorption and scattering by phytoplankton particles, inorganic and
organic particles, and CDOM.
4
Standard algorithms for the estimation of chl-a concentrations in the open ocean
rely on reflectances in the blue and green spectral regions (Gordon and Morel, 1983;
Gordon et al., 1988; O’Reilly et al., 1998; O’Reilly et al., 2000). In inland and coastal
waters, the reflectance in these spectral regions is highly affected by the combined
absorption by phytoplankton pigments, particles, and CDOM and is seldom correlated
with the concentration of phytoplankton particles. Consequently, the algorithms typically
applied for the remote estimation of chl-a concentrations in oceanic waters are ineffective
in many inland and coastal waters and researchers have directed their efforts towards the
development of different algorithms for the remote estimation of chl-a concentrations in
these waters (Neville and Gower, 1977; Gower, 1980; Gitelson et al., 1985; Dekker,
1991; Mittenzwey et al., 1991; Gitelson, 1992; Gons, 1999; Gower et al., 1999; Gons et
al, 2000; Kallio et al., 2001; Strömbeck and Pierson, 2001; Gons et al., 2002, Gons et al.,
2008).
Studies of the optical properties of inland and coastal waters revealed the potential
of the red and near infrared (NIR) spectral regions, where the reflectance is less affected
by the presence of particles and CDOM, for the remote estimation of chl-a concentrations
and initiated the development of algorithms based on these spectral regions. These
algorithms take advantage of several spectral features related to chl-a in the red and NIR
spectral regions.
Some of the algorithms are based on quantification of the reflectance (ρ) peak at
685 nm caused by solar-induced chl-a fluorescence (Neville and Gower, 1977; Gower,
1980; Gower and Borstad, 1990). The increase in fluorescence with an increase in chl-a
concentration made it possible to use the peak magnitude at 685 nm above a baseline
5
from 650 to 730 nm, the fluorescence line height (FLH), for the estimation of chl-a
concentrations (Gower, 1980). Today the estimation of chl-a concentrations through the
FLH is typically applied in coastal waters and parts of the open ocean (Gower and King,
2007; Ryan et al., 2009) and the FLH was only recently evaluated for the detection of
phytoplankton blooms in inland waters (Binding et al., 2011a). Disadvantages of the FLH
for the estimation of chl-a concentrations include its sensitivity to the natural variability
of the chl-a specific absorption coefficient of phytoplankton, a*φ(λ), the re-absorption of
fluoresced light by chl-a, and the natural variability of the quantum yield of chl-a
fluorescence (Babin et al., 1996).
Many algorithms for the remote estimation of chl-a concentrations in inland and
coastal waters are based on the reflectance peak around 700 nm, caused by a minimum in
the combined absorption by phytoplankton pigments and water (Vasilkov and
Kopelevich, 1982; Gitelson, 1992; Yacobi et al., 1995; Han and Rundquist, 1997), and
the reflectance trough at 675 nm, which is related to the chl-a absorption maximum in the
red region. Some algorithms explored the ratio of the reflectance at the peak around 700
nm to the reflectance at the peak around 560 nm, which is related to a minimum in the
combined absorption by phytoplankton pigments, particles, and CDOM. Close
relationships, with Pearson correlation coefficients (r) of 0.91 to 0.99, were found
between ρ(700)/ρ(560) and chl-a concentration for the rivers Don and Donec in Russia,
the Azov Sea, and Lake Balaton in Hungary (Gitelson et al., 1986).
Alternative algorithms included the position and magnitude of the reflectance
peak around 700 nm (Gitelson and Kondratyev, 1991; Gitelson, 1992; Gitelson et al.,
1993), the reflectance ratios ρ(705)/ρ(675) and ρ(705)/ρ(670), (Gitelson et al., 1985;
6
Mittenzwey et al. 1991; Quibell, 1991; Mittenzwey et al., 1992; Dekker, 1993; Han and
Rundquist, 1997), and some more complex empirical algorithms (Mittenzwey et al.,
1991).
Models for the remote estimation of chl-a concentrations are based on a
fundamental relationship (Gordon, 1973; Gordon et al., 1975) of remote sensing
reflectance, ρrs(λ), and the inherent optical properties of water – the absorption
coefficient, a(λ), and the backscattering coefficient, bb(λ):
ρrs ( λ) =
bb ( λ)
f ( λ)
Q( λ) a ( λ) + bb ( λ)
(1.1)
where f(λ) describes the sensitivity of the reflectance to variations in the solar zenith
angle (Morel and Gentili, 1991) and Q(λ) expresses the bidirectional properties of the
reflectance (Morel and Gentili, 1993).
Gons (1999) reformulated the ratio ρ(705)/ρ(675) in terms of a(λ) and bb(λ) for
subsurface irradiance reflectance spectra and partitioned a(λ) into the absorption
coefficients of phytoplankton, aφ(λ), non-algal particles, aNAP(λ), CDOM, aCDOM(λ), and
water, aw(λ), to develop a semi-empirical model for the remote estimation of chl-a
concentrations in inland and coastal waters (Gons, 1999, Gons et al., 2000) in the form:
[Chl - a] =
ρ (λ 2 )
p
(a w (λ 2 ) + bb ) - a w (λ1 ) - bb
ρ (λ 1 )
*
aφ (λ 1 )
(1.2)
where ρ(λ1) is the reflectance at 672 nm, a wavelength close to the chl-a absorption
maximum in the red region, ρ(λ2) is the reflectance at 704 nm, aw(λ1) and aw(λ2) are the
7
absorption coefficients of water at 672 and 704 nm, bb is the backscattering coefficient,
which is assumed to be wavelength independent, a*φ(λ1) is the chl-a specific absorption
coefficient of phytoplankton at 672 nm, and p is a coefficient introduced to improve the
fit of the model.
This model is based on several simplifications: (a) the absorption at around 672
nm is controlled by chl-a and water and is minimally affected by the rest of the
constituents, (b) the absorption by pigments, particles, and CDOM at around 704 nm is
insignificant compared to the absorption by water, and (c) bb is wavelength independent
and can be derived from the NIR wavelengths. The coefficients for a*φ(λ1) and p were
retrieved through regression for datasets from several inland waters.
Slightly different wavebands were applied in an algorithm for the remote
estimation of chl-a concentrations in inland and coastal waters for the spectral bands of
the MEdium Resolution Imaging Spectrometer (MERIS) presented by Gons et al. (2002).
This algorithm was revised in 2005 due to a shift in the originally planned wavebands of
the satellite (Gons et al., 2005) and was of the final form:
[Chl - a] =
ρ (709)
p
(a w (709) + bb ) - a w (665) - bb
ρ (665)
*
aφ (λ1 )
(1.3)
where ρ(665) is the reflectance for MERIS band 7 (centered at 665.00 nm), ρ(709) is the
reflectance for MERIS band 9 (centered at 708.75 nm), aw(665) and aw(709), the
absorption coefficients of water for MERIS bands 7 and 9, were set to 0.4 and 0.7, bb, the
backscattering coefficient, was derived from MERIS band 12 (centered at 778.75 nm), p
was set to 1.06, and a*φ(665) was 0.0161.
8
To estimate bb from (Eq. 1.1) an experimental value of 0.082 (Gons, 1999) was
used to represent the ratio f(λ)/Q(λ) and a conversion factor of 0.60 was used to relate the
reflectance of the MERIS level-2 standard product to bb:
bb =
1.61ρ (779)
0.082 − 0.6ρ (779)
(1.4)
where ρ(779) is the reflectance for MERIS band 12 (centered at 778.75 nm).
Dall’Olmo et al. (2003) applied a conceptual model, originally developed for the
estimation of the chl-a content in terrestrial vegetation (Gitelson et al., 2003), to turbid
productive waters:
[Chl-a] ∝ [ρrs-1(λ1) - ρrs-1(λ2)] × ρrs(λ3)
(1.5)
where ρrs-1(λ1) is considered maximally sensitive to absorption by chl-a, even though it is
still affected by absorption and scattering by the rest of the constituents, and ρrs-1(λ2) is
minimally sensitive to absorption by chl-a. If the sensitivities of ρrs-1(λ2) and ρrs-1(λ1) to
the presence of NAP and CDOM are comparable, it is possible to remove the effects of
the absorption by these constituents through the subtraction of ρrs-1(λ2) from ρrs-1(λ1). The
multiplication of [ρrs-1(λ1) – ρrs-1(λ2)] with ρrs(λ3) should remove the effects of the
variability in bb(λ) on the chl-a estimation if ρrs(λ3) is minimally sensitive to absorption
by chl-a, NAP, and CDOM and representative of bb(λ). The optimal wavelengths for the
estimation of chl-a concentrations from 2 to 180 mg m-3 were identified at λ1 = 670 nm,
λ2 = 710 nm, and λ3 = 740 nm (Dall’Olmo and Gitelson, 2005). The three-band model
relies, similar to Gon’s model (Gons, 1999), on a wavelength independent bb in the
spectral range from λ1 to λ3.
9
For waters without high concentrations of NAP and CDOM it is possible to
disregard the subtraction of ρs-1(λ2) and this special case corresponds to a two-band model
developed by Stumpf and Tyler (1988):
[Chl-a] ∝ ρrs-1(λ1) × ρrs(λ3)
(1.6)
which is fundamentally different from the widely applied reflectance ratio:
[Chl-a] ∝ ρrs-1(λ1) × ρrs(λ2)
(1.7)
where ρrs-1(λ1) is the inverse reflectance around the chl-a absorption maximum in the red
region at 675 nm and ρrs (λ2) represents the reflectance peak around 705 nm.
Even though the three-band model (Eq. 1.5) and two-band models (Eq. 1.6 and
1.7) are potentially susceptible to the natural variability of the chl-a specific absorption
coefficient of phytoplankton and the quantum yield of chl-a fluorescence, they were
successfully applied for the remote estimation of chl-a concentrations with field
spectrometers (Gitelson et al., 2008; Gitelson et al., 2011; Yacobi et al., 2011, Gurlin et
al., 2011), hyperspectral aerial imagery (Moses et al., 2012a), satellites (Moses et al.,
2009a,b; Moses et al., 2012b), and modeled reflectance spectra (Gilerson et al., 2010).
10
1.3
Remote estimation of phycocyanin concentrations in inland
waters
The presence of cyanobacteria in freshwater ecosystems is a typical consequence
of eutrophication and the adverse effects of cyanobacterial blooms on humans and the
environment include surface scums, a foul odor and taste of water intended for human
consumption, health risks for humans, livestock, and wildlife, and fish kills in
consequence of oxygen depletion and the release of cyanotoxins. Many species of
cyanobacteria produce several types of water soluble cyanotoxins (Carmichael, 1997).
These include a variety of hepatotoxic microcystins, the hepatotoxic nodularin, the
cytotoxic cylindrospermopsin, and the neurotoxic anatoxin, β-Methylamino-L-alanine
(BMAA), and saxitoxins (Carmichael, 1997), which are typically released when
cyanobacterial blooms die or cyanobacteria are ingested (Carpenter et al., 1998). The
concentration of C-phycocyanin, a pigment exclusively found in freshwater
cyanobacteria, is a measure of the cyanobacterial productivity of inland waters and
provides invaluable insights in the environmental health of these ecosystems.
Some algorithms for the remote estimation of phycocyanin concentrations in
inland waters (Schalles and Yacobi, 2000; Ruiz Verdú et al., 2008) explored the ratio of
the reflectance at a smaller peak around 650 nm to the reflectance at a trough around 620
nm and the reflectance ratio ρ(650)/ρ(625) was an effective predictor of phycocyanin
concentrations in Carter Lake, in western Iowa, USA (Schalles and Yacobi, 2000). The
algorithms typically applied for the remote estimation of phycocyanin concentrations in
inland waters are based on the reflectance peak around 700 nm, caused by the minimum
11
of the combined absorption by phytoplankton pigments and water (Vasilkov and
Kopelevich, 1982; Gitelson, 1992; Yacobi et al., 1995; Han and Rundquist, 1997), and
the reflectance trough at 620 nm, which is related to the C-phycoyanin absorption
maximum in the red region (Glazer et al, 1973). The reflectance ratios ρ(710)/ρ(620) and
ρ(705)/ρ(620) were successfully applied for the remote estimation of phycocyanin
concentrations from hyperspectral aerial imagery in shallow mesotrophic and eutrophic
inland waters in the UK (Hunter et al., 2009; Tyler et al., 2009, Hunter et al., 2010).
Practically the same reflectance ratio was applied in a semi-empirical model,
developed by Simis et al. (2005) for the spectral bands of the MERIS satellite sensor. The
estimation of the phycocyanin concentrations with this model includes the application of
the reflectance ratios ρ(709)/ρ(665) and ρ(709)/ρ(620) in form of several nested algorithms.
Firstly, the absorption coefficient of chl-a at 665 nm, achl-a(665), is estimated to
correct for possible absorption by chl-a in the spectral region of interest for the estimation
of phycocyanin:
  ρ (709) 


achl−a (665) =  
× [(a w (709) + bb )] - a w (665) - bb  × γ −1

  ρ (665) 




(1.8)
where ρ(665) is the simulated reflectance for MERIS band 7 (centered at 665.00 nm),
ρ(709) is the simulated reflectance for MERIS band 9 (centered at 708.75 nm), aw(665)
and aw(709), the absorption coefficients of water for MERIS bands 7 and 9, were set to
0.4 and 0.7 (from Buiteveld, 1994), bb, the backscattering coefficient, was derived from
MERIS band 12 (centered at 778.75 nm), (Eq. 1.4), and γ, a correction factor to correct
for differences in the estimated and analytically measured aφ(665), was set to 0.68. This
12
algorithm represents a slight modification of Gons’ algorithm (Gons et al., 2005), (Eq.
1.3), and relies on the same simplifications.
Secondly, the absorption coefficient of phycocyanin at 620 nm, aPC(620), is
estimated where the conversion factor ε is introduced to relate the absorption by chl-a at
665 nm to its absorption at 620 nm:
  ρ (709) 


a PC (620) =  
× [(a w (709) + bb )] - a w (620) - bb  × δ −1 − [ε × achl-a (665)]

  ρ (620) 




(1.9)
where ρ(620) is the simulated reflectance for MERIS band 6 (centered at 620.00 nm),
ρ(709) is the simulated reflectance for MERIS band 9 (centered at 708.75 nm), aw(620)
and aw(709), the absorption coefficients of water for MERIS bands 6 and 9, were set to
0.3 and 0.7 (from Buiteveld, 1994), bb was derived from MERIS band 12 (centered at
778.75 nm), δ, a correction factor to correct for differences in the estimated and
analytically measured aφ(620), was set to 0.84, and ε was set to 0.24.
The concentration of phycocyanin is retrieved through the division of aPC(620) by
the phycocyanin specific absorption coefficient of phytoplankton, a*PC(620):
[PC] =
a PC (620)
∗
a PC
(620)
(1.10)
where a*PC(620) was set to 0.0095 m2 mg-1.
These algorithms were successfully applied for the remote estimation of
phycocyanin concentrations in European and North American inland waters with field
spectrometers and from hyperspectral aerial imagery (Simis et al., 2005; Simis et al.,
2007; Ruiz Verdú et al., 2008; Randolph et al. 2008; Hunter et al., 2010).
13
1.4
Objective
The overall objective of this dissertation is to investigate the performance of NIR-
red models for the remote estimation of chl-a and phycocyanin concentrations in
optically complex turbid productive waters.
Specifically, the research presented in this dissertation shows the
(1) Importance of the red and NIR spectral regions for the remote estimation of chl-a
concentrations in turbid productive waters,
(2) Calibration and validation of three NIR-red models for the remote estimation of chl-a
concentrations up to 25 mg m-3, typical for many inland and coastal waters, for
independent and statistically different datasets, for the spectral bands of the MERIS
and MODIS satellite sensors,
(3) Potential of a simple two-band NIR-red algorithm for the MERIS satellite sensor to
accurately estimate chl-a concentrations in comparison with several more complex
recently published NIR-red algorithms,
(4) Investigation of a complex NIR-red algorithm for the remote estimation of
phycocyanin concentrations in inland waters with a diverse species composition of
the phytoplankton, and
(5) Development of NIR-red algorithms for the remote estimation of chl-a concentrations
from atmospherically corrected multi-temporal AISA imagery.
14
1.5
Structure
Chapter 1 discusses the influence of cultural eutrophication on the productivity of
inland and coastal waters, the importance of the chl-a concentration in studies of the
productivity of these waters, and the application of NIR-red models for the remote
estimation of chl-a and phycocyanin concentrations.
Chapter 2 provides a description of the datasets and methods. It includes a
description of the study areas at the Fremont Lakes and Branched Oak Lake in eastern
Nebraska, USA and it presents details of the measurements of the bio-physical and biooptical water quality parameters collected in the field and laboratory, the calibration and
validation of the NIR-red models for the estimation of chl-a concentrations in optically
complex turbid productive waters, and the application of the NIR-red models for the
estimation of chl-a concentrations from multi-temporal hyperspectral aerial imagery.
Chapter 3 presents a detailed description of the bio-physical and bio-optical water
quality parameters measured in the Fremont Lakes and Branched Oak Lake in eastern
Nebraska, USA, in comparison to water quality parameters typical for inland and coastal
waters around the world.
Chapter 4 discusses the effects of the presence of NAP and CDOM, which make
the algorithms typically applied for oceanic waters ineffective for optically complex
inland and coastal waters, for datasets collected at the Fremont Lakes in 2008 and 2009,
and shows the importance of the red and NIR spectral regions for the development of
algorithms for the remote estimation of chl-a concentrations in inland and coastal waters.
The performance of three NIR-red models for the remote estimation of chl-a
15
concentrations, calibrated and validated for independent and statistically different
datasets, is investigated and the potential of a simple two-band NIR-red algorithm for the
MERIS satellite sensor to accurately estimate chl-a concentrations in comparison with
several more complex recently published NIR-red algorithms is shown. Possible
constraints for the application of the developed algorithms for the estimation of chl-a
concentrations in turbid highly productive inland waters are investigated for the dataset
collected at Branched Oak Lake in summer 2011. The performance of a NIR-red
algorithm (Eqs. 1.8 to 1.10) for the remote estimation of phycocyanin in inland waters
with a diverse phytoplankton species composition is investigated for a part of the dataset
collected at the Fremont Lakes in 2009.
Chapter 5 shows the application of NIR-red models for the remote estimation of
chl-a concentrations from atmospherically corrected multi-temporal AISA imagery. The
models were calibrated for the stations from two images and validated for the stations
from three different images collected in summer and fall 2008. This chapter discusses the
effects of the atmospheric correction on the remote estimation of the chl-a concentrations
and the sensitivity of the developed algorithms to variations in the chl-a specific
absorption coefficient of phytoplankton.
Chapter 6 presents the conclusions for the investigation of NIR-red models for the
remote estimation of chl-a concentrations in optically complex turbid productive waters
and discusses potential challenges for development of a universally applicable NIR-red
algorithm for the remote estimation of chl-a concentrations in these waters.
16
1.6
Publications
Moses, W. J., Gitelson, A.A., Perk, R. L., Gurlin, D. Rundquist, D. C., Leavitt, B. C.,
Barrow, T. M., and P. Brakhage (2012). Estimation of chlorophyll-a concentration in
turbid productive waters using airborne hyperspectral data, Water Research, 46: 9931004.
Gurlin, D., Gitelson, A.A., and W.J. Moses (2011) Remote estimation of chl-a
concentration in turbid productive waters - Return to a simple two-band NIR-red
model? Remote Sensing of Environment, 115, 3479-3490.
Gitelson, A.-A., Gurlin, D., Moses, W., & Yacobi, Y. (2011) Remote estimation of
chlorophyll-a concentration in inland, estuarine, and coastal waters. In: Q. Weng
(Ed.) Advances in environmental remote sensing: sensors, models, and applications
(pp. 449-478), Boca Raton: CRC Press, Taylor and Francis Group.
Gilerson, A., Gitelson, A.-A., Zhou, J., Gurlin, D., Moses, W.-J., Ioannou, I., and Ahmed,
S.-A. (2010) Algorithms for remote estimation of chlorophyll-a in coastal and inland
waters using red and near infrared bands. Optics Express, 18, 24109-24125.
Gitelson, A.A., Gurlin, D., Moses, W.J., and Barrow, T., (2009). A bio-optical algorithm
for the remote estimation of the chlorophyll-a concentration in case 2 waters,
Environmental Research Letters, 4, Article ID: ERL/310739/SPE
Gitelson, A.-A., Dall’Olmo, G., Moses, W., Rundquist, D., Barrow, T., Fisher, T.-R.,
Gurlin, D., and J. Holz (2008) A simple semi-analytical model for remote estimation
of chlorophyll-a in turbid productive waters: Validation. Remote Sensing of
Environment, 112, 3582-3593.
17
Chapter 2. Datasets and Methods
2.1
Study area
Field data were collected at 152 stations at the Fremont Lakes in eastern Nebraska,
USA, from summer 2008 through summer 2009 (Fig. 2.1) and at 40 stations at Branched
Oak Lake in eastern Nebraska, USA, in summer 2011. The field data included a set of
water quality parameters and hyperspectral reflectance measurements.
Norfolk
Scottsbluff
(
!
(
!
Nebraska
Columbus Fremont
(
!
(
!
North Platte
-
Omaha
(
!
(
!
Lincoln
(
!
(
!
Kearney
( Hastings
!
(
!
0
100
200
Kilometers
AISA Image of the Fremont Lakes
taken on 11/19/2011
07#
#
* *
04
#
*
#
*
05
03
#
*
02
#
*
01
#
*
16
#
*
#
*
0
1
20
17
#
*
18 #
*
VL
2
Kilometers
Figure 2.1 Map of the study area at the Fremont Lakes in eastern Nebraska, USA. The
positions of stations are indicated with the red markers.
18
The Fremont Lakes are part of the Fremont Lakes State Recreation Area, situated
around 5 km west of Fremont, Nebraska, in the Platte River valley. They currently
comprise around 121 ha of surface water and are of high recreational value for thousands
of people in eastern Nebraska (Fig. 2.2). The relatively shallow lakes were a result of
sand and gravel mining operations in the 1930s and water is mainly supplied through
groundwater inflows.
Figure 2.2 Fremont Lake 07 in summer 2008, photo by B. Leavitt, CALMIT, UNL.
The water quality in the Fremont Lakes is monitored by the Nebraska Department
of Environmental Quality (NDEQ) and high concentrations of total phosphorus, total
nitrogen and chl-a are found in several of the Fremont Lakes (NDEQ Water Quality
Division, 2010). From 2004 to 2006, one of the most popular lakes for recreational
activities was closed for 25 weeks due to the presence of algal toxins. The lake was
treated with aluminum sulfate in 2007 to reduce the phosphorus concentration and
productivity of the lake. In 2008, the productivity was significantly reduced and the lake
19
contained only moderate chl-a concentrations at the time of field data collection. The
highly variable bio-physical and bio-optical conditions found in the Fremont Lakes are
typical for many inland and coastal waters and make these lakes ideal for the
development of algorithms for the remote estimation of chl-a concentrations in turbid
productive waters.
Branched Oak Lake is part of the Branched Oak State Recreation Area, situated
around 6 km west of Raymond, Nebraska. It is one of 10 reservoirs in the Salt Creek and
Tributaries Flood Control Project by the U.S. Army Corps of Engineers and provides
flood control for the City of Lincoln, Nebraska. Branched Oak Lake comprises 728 ha of
surface water and supports a multitude of recreational activities. It provides ideal
conditions for watersports and attracts thousands of outdoor enthusiasts on a typical
summer weekend.
The water quality of Branched Oak Lake is compromised by sediments, nutrients
and phytoplankton and problems related to total phosphorus, total nitrogen, and chl-a
concentrations are apparent (NDEQ Water Quality Division, 2010). The presence of
moderate-to-high chl-a concentrations and some algal toxins is typical in the summer
months. The bio-physical and bio-optical conditions found in Branched Oak Lake are
typical for many inland waters and make it possible to evaluate algorithms for the remote
estimation of chl-a concentrations for turbid highly productive inland waters.
20
2.2
Field measurements
2.2.1 Bio-physical and bio-optical water quality parameters
The set of water quality parameters measured in the field and laboratory included
the concentration of dissolved oxygen, water temperature, electrical conductivity, the
concentrations of the phytoplankton pigments chl-a, -b, and -c, the concentration of
phycocyanin for 15 stations in 2009, the concentration of suspended solids, and Secchi
disk depth and turbidity. Firstly, dissolved oxygen, water temperature, and electrical
conductivity were measured with an YSI 556 Multiprobe System (YSI, Inc.) and salinity
was automatically calculated from water temperature and electrical conductivity.
Secondly, surface water samples for the laboratory analysis of the chl-a, -b, and -c
concentrations, phycocyanin concentrations, suspended solids, and the absorption
coefficients ap(λ), aNAP(λ), aφ(λ), and aCDOM(λ) were collected at a depth of 0.5 m and
stored in a dark and cool container for subsequent filtration in the laboratory. Finally,
water transparency was measured with a standard Secchi disk, and turbidity was
measured with a HACH 2100 portable turbidimeter.
The water samples for the laboratory analysis of the chl-a, -b, and -c
concentrations and the absorption coefficients ap(λ), aNAP(λ), and aφ(λ) were filtered
through 25 mm Whatman GF/F filters within 24 hours after collection. We filtered
volumes of 25 and 50 mL of water to measure the concentration of chl-a fluorometrically
and volumes of 50 to 3000 mL of water to measure the concentrations of chl-a, -b, and –c
spectrophotometrically (Tab. 2.1). Sometimes multiple filters were required to collect
sufficient volumes of phytoplankton particles to reach the detection limits of the
21
algorithms for the calculation of the chl-a, -b, and -c concentrations (Ritchie, 2008).
Filtration volumes from 25 to 750 mL were typically sufficient to reach the optimal
optical densities of particles on the filters (Mitchell, 1990) for the laboratory analysis of
ap(λ), aNAP(λ), and aφ(λ). The filters for the extraction of chl-a, -b, and -c were stored in a
freezer at a temperature of -18°C for a maximum of four weeks to prevent the
degradation of the pigments and the absorption coefficients were analyzed shortly after
the filtration of the water samples.
We filtered the water samples for the laboratory analysis of phycocyanin through
25 mm Whatman GF/C filters to collect sufficient volumes of phytoplankton particles in
conditions with low-to-moderate phycocyanin concentrations. These filters retained the
relatively large sized Cyanobacteria typically found in inland waters effectively and made
it possible to filter volumes of 150 to 500 mL of water. The filters for the extraction of
phycocyanin were stored in the freezer and shipped to Indiana University-Purdue
University Indianapolis (IUPUI) on dry ice for the analysis of the phycocyanin
concentrations after the field data collection was completed in 2009.
The water samples for laboratory analysis of total suspended solids (TSS),
inorganic suspended solids (ISS), and organic suspended solids (OSS) were filtered
through 47 mm Whatman GF/F filters within 42 hours after collection. We filtered a
volume of 150 to 2250 mL, estimated through the turbidity of the station (Fig. 2.3), for
each station (Tab. 2.1) and 150 mL of the filtrates were filtered through 47 mm Whatman
0.2 µm nylon membranes for the laboratory analysis of aCDOM (λ).
22
Table 2.1 Filtration volumes for the laboratory analysis of the chl-a, -b, and –c
concentrations, phycocyanin concentrations, absorption coefficients, and suspended
solids for 06/15/2009.
Lake
Chl-a
Chl-a, -b, and -c
mL
mL
FSL01
FSL02
FSL03
FSL05
FSL07
FSL16
FSL17
FSL20
Victory Lake
50
50
50
50
50
50
50
50
50
3 × 500
3 × 500
3 × 400
1 × 300
3 × 100
3 × 100
3 × 250
3 × 500
3 × 150
Number of
replicates
3
3
PC
Absorption
coefficients
mL
TSS
300
150
500
500
400
150
100
100
250
500
150
1500
1500
1200
900
300
300
750
1500
450
2
1
1
mL
Chl-a – Fluorometrically measured chlorophyll-a concentration, chl-a, -b, and –c – spectrophotometrically
measured chlorophyll-a, -b, and –c concentrations, PC – phycocyanin concentration, TSS – concentration
of total suspended solids
2.5
R² = 0.61
2.0
Vf, L
1.5
1.0
0.5
0.0
0
5
10
15
20
25
Turbidity, NTU
Figure 2.3 The turbidity provided a first estimate of the filtration volumes (Vf) for the
TSS measurements (Vf shown for lakes 01, 02, 03, 07, and 16 in 2008 and 2009).
23
Volume scattering coefficients, β(θ, λ), were measured in 2009 with a customized
ECO Triplet sensor (WET Labs, Inc.) at wavelengths of 570, 660, and 740 nm and
corrected for salinity, aw(λ), ap(λ), and aCDOM(λ). The values for aw(λ) were taken from
Mueller (2003) and the values for ap(λ) and aCDOM(λ) from the laboratory measurements.
The volume scattering of particles, βp(117°, λ), and volume scattering of water, βw(117°,
λ) were derived from the corrected volume scattering coefficients, β(117°, λ), to calculate
the particulate backscattering coefficients, bbp(λ).
The backscattering coefficients, bb(λ), were calculated from the sum of bbp(λ) and
the backscattering coefficients of pure water, bbw(λ), (WET Labs, Inc., 2008). We
calculated the backscattering coefficients, bb(λ), for 44 stations in 2009 due to the
saturation of the sensor at 18 stations and a failure of the sensor at one station.
2.2.2 Reflectance
Hyperspectral reflectance data were collected with two inter-calibrated Ocean
Optics® USB2000 spectrometers in areas of optically deep water in the spectral range
from 400 to 900 nm with a spectral resolution of ~ 0.3 nm. One of the spectrometers was
connected to a 25° field-of-view optical fiber. The optical fiber was taped to a 2-m long,
hand-held dark blue pole and the tip of the fiber was kept just beneath the water surface
to measure the below-surface upwelling radiance, Lu(λ), at nadir on the sun-lit side of the
boat. The second spectrometer was connected to an optical fiber fitted with a cosine
collector to create a 180° field-of-view. The optical fiber was mounted on a mast at the
highest possible point on the boat to measure the incident irradiance, Ed(λ), at zenith
concurrently with upwelling radiance. The measurements were collected from 10 a.m. to
24
2 p.m. with widely variable solar zenith angles from 66.44° in fall 2008 to 18.16° in
spring 2009. Inter-calibration of the instruments, to account for potential differences in
their transfer functions, was achieved through synchronized measurements of the
upwelling radiance, Lref(λ), from a white Spectralon® reflectance standard (Labsphere,
Inc.), and the irradiance incident on the standard, Eref(λ). The remote sensing reflectance
was computed by:
ρrs (λ) =
Lu (λ) E ref (λ)
ρref (λ) t
×
× 100 ×
× 2 × F (λ )
Ed (λ) Lref (λ)
π
n
(2.1)
where ρref(λ) is the irradiance reflectance of the Spectralon® reflectance standard, π is
introduced to transform the irradiance reflectance ρref(λ) into a radiance reflectance, t is
the water-to-air transmittance (t = 0.98, Preisendorfer, 1986), n is the refractive index of
water relative to air (n = 1.33 at 20°C), and F(λ) is the spectral immersion factor (Ohde
and Siegel, 2003). The spectra were processed in real time with the software CDAP,
developed at CALMIT, University of Nebraska – Lincoln, and the median of at least six
spectra collected at each station was considered representative for the station.
The hyperspectral reflectance data for several stations at Branched Oak Lake in
summer 2011 were collected with a single spectrometer due to an instrument failure of
the second spectrometer. Reference measurements of Lref(λ) were collected at each station
to account for variations in Ed(λ) on this sunny day with clear sky conditions.
25
2.3
Laboratory measurements
Chl-a was extracted in subdued light conditions in a laboratory at the University
of Nebraska – Lincoln. Therefore the filters were covered with 10 mL 99.5% ethanol
(459844-2L ACS reagent from Sigma Aldrich) and subjected to a five minute heat
treatment at a temperature of 78°C (modified from Nusch, 1980). The extractions
continued for four hours at a temperature of 1 – 4°C and the samples were centrifuged for
5 minutes in a Cole-Parmer® EW-17250-10 fixed-speed centrifuge to remove any filter
particles from the extracts.
We decided to extract chl-a in ethanol for practical and safety reasons. Even
though 99.5% ethanol is a highly flammable liquid and requires precautionary procedures
in the laboratory it imposes less of a health risk compared to the more widely used
acetone and is an efficient extractant if resistant phytoplankton species, for example
chloropthyta, are present (Nusch, 1980). Its compatibility with polystyrene made it
possible to centrifuge the extracts in standard polystyrene centrifuge tubes (Ritchie,
2008).
The chl-a concentrations were measured fluorometrically with a Turner 10-AU005 CE fluorometer (Turner Designs, Inc.) set up with a Blue Mercury Vapor Lamp and
a 436-680 nm excitation-emission filter combination (Optical Kit 10-040R). This
combination of interference filters results in maximum sensitivity to chl-a and reduces
interferences from chl-b and phaeophytins typically found in inland water samples
(Welshmeyer, 1994).
26
The fluorometrically measured chl-a concentrations of the extracts were corrected
for the dilution factor, the volume of the extractant, and the volume of water filtered by:
Chl - a =
Chl - a in
extract
× DF × Ve
Vf
(Eq. 2.2)
where Chl-ain extract is the chl-a concentration of the extract in mg m-3, DF represents the
dilution factor, Ve is the volume of extractant in m-3, and Vf is the volume of water filtered
in m-3. The chl-a concentrations for three sub-samples (Tab. 2.1) were averaged to
calculate the final chl-a concentration representative of each station.
We calibrated the fluorometer every three months with a 100 mg m-3 chl-a
standard from Anacystis nidulans (C6144-1MG from Sigma Aldrich). Therefore the chl-a
was dissolved in 1 L 99.5% ethanol and the absorbance of the standard was measured
with a Cary 100 spectrophotometer (Agilent Technologies, Inc.). The chl-a concentration
of the standard was calculated from the trichromatic equation published by Ritchie
(2008). Standard curves to study the linearity of the single point calibration of the
fluorometer were prepared at the time of calibration and included concentrations of 0,
0.5, 1, 2.5, 5, 10, 25, 50, 100, and 200 mg m-3 (Fig. 2.4).
The stability of the calibration was studied through measurements of the
fluorescence of solid secondary standards for chl-a analysis (Turner Designs, Inc.). Even
though the fluorometer readings for these standards were relatively stable for three
months (Fig. 2.5) differences of up to 1.4 mg m-3 for the “high” chl-a standard and up to
0.2 mg m-3 for the “low” chl-a standard were revealed when the fluorometer readings
from 09/03/2008 to 11/24/2008 were compared to the readings from the day the
instrument was calibrated (Fig. 2.6).
27
To compensate for these differences standard curves for chl-a concentrations from
0 to 200 mg m-3 were prepared each time chl-a concentrations were measured in 2009
and 2011. This resulted in a significant improvement in the relationship of chl-a
concentration and aφ(675) from 2008 to 2009 (Fig. 3.7).
Chl-a - Standard, mg m-3
200
150
100
50
R² = 0.99994
0
0
50
100
Chl-a - 10-AU, mg
150
200
m-3
Figure 2.4 Standard curve from 08/26/2008 for chl-a concentrations from 0 to 200 mg m-3.
Fluorometer Reading, mg m-3
Chlorophyll-a Solid Standards
36
High
30
24
18
12
6
Low
0
08/20 09/03 09/17 10/01 10/15 10/29 11/12 11/26
Date
Dates of chl-a measurements in 2008
Figure 2.5 Variability of the fluorescence readings of the “high” and “low” solid chl-a
standards after the fluorometer calibration on 08/26/2008.
Fluorometer Reading, mg m-3
28
Chlorophyll-a Solid Standards
High
36
35
34
High
Low
5.4
5.2
5.0
Low
33
4.8
4.6
32
08/20 09/03 09/17 10/01 10/15 10/29 11/12 11/26
Date
Dates of chl-a measurements in 2008
Figure 2.6 Variability of the fluorescence readings of the “high” and “low” solid chl-a
standards after the fluorometer calibration on 08/26/2008 shown in more detail.
Chl-a, -b, and –c were extracted in 7.5 mL 99.5% ethanol in 2008 and 2009 and in
10 mL 99.5% ethanol in 2011 with the same technique applied for chl-a. The absorbance
of the extracts was measured with the Cary 100 spectrophotometer and the chl-a, -b, and
–c concentrations were calculated from the trichromatic equations published by Ritchie
(2008) followed by Eq. 2.2. The chl-a concentrations for three sub-samples (Tab. 2.1)
were averaged to calculate the final chl-a concentration representative of each station.
Phycocyanin was extracted in 25 mL 50 mM phosphate buffer (Sarada et al.,
1999) in a laboratory at IUPUI. The phycocyanin concentrations were measured with a
TD-700 fluorometer (Turner Designs, Inc.) set up with a Cool White Mercury Vapor
Lamp and a 630-660 nm excitation-emission filter combination (Optical Kit 10-305). The
fluorometer was calibrated with highly purified c-phycocyanin from Spirulina sp.
(P6161-.5MG from Sigma-Aldrich).
29
Details of the extraction procedure and instrument calibration are found in Li et
al. (2010) and Nguy-Robertson (2012). The phycocyanin concentrations for two
subsamples were averaged to calculate the final concentration for each station.
The concentrations of total suspended solids (TSS), inorganic suspended solids
(ISS), and organic suspended solids (OSS) were measured gravimetrically (Eaton et al.,
2005). Therefore the samples were dried for 24 hours at a temperature of 103-105⁰C to
measure TSS (Fig. 2.7) and ignited for one hour at a temperature of 550⁰C to separate the
suspended solids in ISS and OSS.
FSL16
Vf = 150 mL
FSL02
Vf = 200 mL
FSL04
Vf = 300 mL
FSL05
Vf = 800 mL
FSL03
Vf = 800 mL
Figure 2.7 The distinctive colors of the dried filters indicate substantial differences in the
composition of the suspended solids for these five lakes on 10/09/2008.
Particulate absorption coefficients were analyzed with the quantitative filter
technique (Mitchell et al., 2003). The measurements of the optical density of the particles
retained on the filters were made shortly after the filtration of the water samples with a
Cary 100 spectrophotometer in the range from 400 to 800 nm in intervals of 1 nm (Fig.
2.8).
30
The signal from a MilliQ water saturated reference filter was subtracted
automatically from the measurements of the optical density and ap(λ) was calculated by:
a p (λ) =
ln(10)
[0.3893 × [ODfp ( λ ) - ODnull ] + 0.4340 × [ODfp ( λ ) - ODnull ] 2 ]
Vf
A
(Eq. 2.3)
where ODfp(λ) is the optical density of the sample, ODnull(λ) is the average optical density
of the sample from 780 to 800 nm required for the null point correction of the
measurements, Vf is the volume of water filtered in m3, and A is the area of the filter in
m2. This equation includes a quadratic function for the pathlength amplification
correction of the measurements (Cleveland and Weidemann, 1993) derived by Dall’Olmo
(2006) for water samples from lakes and reservoirs in Nebraska and laboratory cultures
of Microcystis and Synechococcus.
The effects of the absorption by pigments were removed after Ferrari and Tassan
(1999). Therefore the samples were treated with 120 µl sodium hypochlorite solution in
MilliQ water (0.1 - 0.2% active Cl) and rinsed with 50 mL MilliQ water after a 20 minute
reaction time. Afterwards aNAP(λ) was measured similarly to ap(λ). The subtraction of
aNAP(λ) from ap(λ) resulted in aφ(λ).
Some of the ODfp measurements for total particulates and NAP for the Branched
Oak Lake 2011 dataset showed instrument errors in the blue spectral region. These scans
were smoothed within an 11 nm window and the ODfp measurements for NAP were
corrected with a polynomial fit (Fig. 2.9).
31
0.5
0.4
ODfp
Total Particulates
0.3
NAP
0.2
0.1
0.0
400
500
600
Wavelength, nm
700
800
Figure 2.8 Measurements of the optical density of the particles retained on the filters for
FSL07 on 06/15/2009. The absorption by pigments was successfully removed through the
sodium hypochlorite solution.
0.75
0.75
ODfp
b 1.00
ODfp
a 1.00
0.50
0.50
0.25
0.25
0.00
0.00
500
600
700
Wavelength, nm
c
0.01
Difference in ODfp
for NAP
400
0.00
400
400
800
500
600
700
500
600
700
Wavelength, nm
800
800
-0.01
Wavelength, nm
Figure 2.9 Measurements of the optical densities of the particles retained on the filters
for a station at Branched Oak Lake on 08/18/2011 (a), after correction (b), and difference
in the optical density for NAP (c).
32
The absorption coefficients of CDOM were measured spectrophotometrically
shortly after the filtration of the water samples. The optical densities of the filtrates were
measured in a 0.1 m cuvette in the range from 200 to 800 nm in intervals of 1 nm with
the Cary 100 spectrophotometer. The signal from a MilliQ water reference sample was
subtracted automatically from the measurements. Filtrates and MilliQ water reference
sample were kept were kept at a constant temperature to minimize a temperature related
absorption feature found around 750 nm (Pegau et al., 1997). The absorption coefficients
of CDOM were calculated by:
a CDOM (λ) =
ln(10)
[ODs (λ) - ODnull ]
l
(Eq. 2.4)
where ODs(λ) is the optical density of the sample, ODnull(λ) is the average optical density
of the sample from 780 to 800 nm for the null point correction, and l is the pathlength of
the cuvette in m. The final absorption coefficients were calculated from an exponential fit
of aCDOM(λ) since the absorption feature at 750 nm was still present in some of the
samples (Bricaud et al, 1981):
aCDOM (λ) = a CDOM (440) - SCDOM ( λ-440)
(Eq. 2.5)
where SCDOM is the spectral slope of CDOM calculated for wavelengths from 350 to 500
nm (Babin et al., 2003).
33
2.4
Calibration and validation of the NIR-red models for the
remote estimation of chl-a concentrations
Three NIR-red models, the three-band model (Eq. 1.5), special case two-band
model (Eq. 1.6), and two-band model (Eq. 1.7), were calibrated and validated for the
spectral bands of the MERIS (Eqs. 1.5 and 1.7) and MODIS (Eq. 1.6) satellite sensors
(Tab. 2.2) to estimate low-to-moderate chl-a concentrations in the Fremont Lakes. These
models were calibrated for the dataset collected in summer and fall 2008 and validated
for the dataset collected in spring and summer 2009.
The accuracy of the chl-a estimations was assessed through the mean absolute
error, MAE, the root mean squared error, RMSE, and the mean normalized absolute
error, MNAE. To compare the absolute estimation errors from the developed NIR-red
algorithms with the errors from several recently published NIR-red algorithms for the
estimation of chl-a concentrations in inland and coastal waters, a one-way analysis of
variance (ANVOA), followed by Tukey’s honestly significant difference multiple
comparison procedure (Tukey’s HSD), was performed in IBM® SPSS® Statistics 20.0
(IBM).
Possible constraints for the application of the developed algorithms for the
estimation of high chl-a concentrations were investigated for the dataset collected in
summer 2011 at Branched Oak Lake.
34
Table 2.2 Spectral bands of the MERIS and MODIS satellite sensors used for the
development of NIR-red algorithms for the remote estimation of chl-a concentrations.
Wavelength
λ1
λ2
λ3
MODIS
MERIS
Band
Center
Width
Band
Center
Width
13
665 nm
662-672 nm
15
748 nm
743-753 nm
7
9
10
665.00 nm
708.75 nm
753.75 nm
660.00-670.00 nm
703.75-713.75 nm
750.00-757.50 nm
The NIR-red model for the estimation of phycocyanin concentrations, developed
by Simis et al. (2005) for the spectral bands of the MERIS satellite sensor (Eqs. 1.8 to
1.10), was applied to estimate phycocyanin concentrations for a small number of stations
from the dataset collected at the Fremont Lakes in 2009. The estimation of the
phycocyanin concentrations with this model involved the application of several
algorithms. We studied the parameterization of these algorithms, which were originally
developed for cyanobacteria dominated inland waters, for the Fremont Lakes 2009
dataset and re-parameterized the algorithms to estimate phycocyanin concentrations in
the Fremont Lakes, for which a mixed species composition of the phytoplankton is more
typical.
35
2.5
Application of NIR-red models for the remote estimation of
chl-a concentrations from multi-temporal AISA imagery
Three NIR-red models, the three-band model (Eq. 1.5), two-band model (Eq. 1.7),
and Gons’ model (Eq. 1.2) were calibrated and validated for five hyperspectral images
with 35 stations from the dataset collected at the Fremont Lakes in 2008. These images
were collected by the hyperspectral airborne imaging spectrometer for applications sensor
(AisaEAGLE), mounted on a Piper Saratoga aircraft, which was flown in an altitude of 3
km above ground at the Fremont Lakes in summer and fall 2008. The images were
acquired on 07/02/2008, 07/14/2008, 09/26/2008, 10/25/2008 and 11/19/2008. These
dates coincided with the field measurements at the Fremont Lakes except for 10/25/2008
where the in situ measurements were taken on 10/24/2008.
The AisaEAGLE sensor is a programmable spectrometer with a maximum of 488
continuous spectral bands with sampling intervals of up to 1.25 nm and a spectral
resolution of 2.9 nm in the spectral range from 400 to 970 nm (Spectral Imaging Ltd.,
2012). The programmable band configurations make it possible to collect data with
application specific spectral characteristics for a multitude of applications from
agricultural precision farming to water quality monitoring.
The hyperspectral images collected at the Fremont Lakes in summer and fall 2008
included 62 continuous spectral bands with sampling intervals from 8.8 nm in the blue
spectral region to 9.6 nm in the NIR spectral region and a spectral resolution of around 10
nm in the spectral range from 409.91 to 981.68 nm. The altitude of 3 km above ground,
36
in which the Piper Saratoga aircraft was flown, resulted in a spatial resolution of the
images of 2 m.
The at-sensor radiances were processed with the default software CaliGeo and the
images were atmospherically corrected with the QUick atmospheric correction (QUAC)
in ENVI 4.8 (Moses et al., 2012), a fully image driven semi-empirical visible-near
infrared through shortwave infrared (VNIR-SWIR) atmospheric correction procedure for
multispectral and hyperspectral imagery (Bernstein et al., 2005a, b; Berstein et al.,
2006). The NIR-red models were applied to the QUAC corrected reflectances after the
dimensionless reflectances were converted to remote sensing reflectances through
division by π. The models were calibrated for the images from 07/14/2008 and
10/25/2008 and validated for the images from 07/02/2008, 09/26/2008, and 11/19/2008.
The accuracy of the chl-a estimations was assessed through the MAE, the RMSE, and the
MNAE.
The statistical analyses presented in chapters three to five of this dissertation were
performed in Microsoft Excel 2007 and 2010 and in IBM® SPSS® Statistics 20.0 (IBM).
37
Chapter 3. Bio-physical and Bio-optical Properties of the
Fremont Lakes and Branched Oak Lake
3.1
Bio-physical properties of the Fremont Lakes
The descriptive statistics of the water quality parameters measured in the Fremont
Lakes in 2008 and 2009 (Tab. 3.1 and 3.2) indicate bio-physical and bio-optical
conditions typical for turbid productive inland (Kallio et al., 2001; Koponen et al., 2002;
Binding et al., 2008; Effler et al., 2010) and coastal (Babin et al., 2003; Binding et al.,
2005; Koponen et al., 2007) waters. The measured chl-a concentrations ranged from 2.3
to 200.8 mg m-3 in 2008 and from 4.0 to 196.4 mg m-3 in 2009 and were practically
identical with the spectrophotometrically measured chl-a concentrations (Fig. 3.1).
200
2008
2009
Chl-a, mg m-3
(Fluorometer)
150
100
50
R² = 0.9884
0
0
50
100
150
Chl-a, mg m-3
(Spectrophotometer)
200
Figure 3.1 Comparison of the spectrophotometrically and fluorometrically measured chl-a
concentrations for the datasets collected at the Fremont Lakes in 2008 and 2009.
38
The distributions of the fluorometrically measured chl-a concentrations were
significantly different from normal distributions in 2008 and 2009 (Shapiro-Wilk test, p <
0.01). They were positively skewed and chl-a concentrations above 50 mg m-3 were
found at only 14 stations in 2008 and 9 stations in 2009 (Fig. 3.2). The extreme values
included stations with chl-a concentrations of 132.4 and 200.8 mg m-3 in one of the lakes
in fall 2008 and a station with a chl-a concentration of 196.4 mg m-3 in the same lake in
early spring 2009.
a 25
2008
Frequency
20
15
10
5
0
0
50
100
150
Chl-a, mg
m-3
200
b 25
250
2009
Frequency
20
15
10
5
0
0
50
100
150
Chl-a, mg
m-3
200
250
Figure 3.2 Frequency distributions of the chl-a concentrations measured in 2008 (a) and
2009 (b).
39
Table 3.1 Descriptive statistics of the water quality parameters measured in 2008.
N
Min
Max
Median
Mean
Standard
Deviation
CV
(%)
(mg m-3)
89
2.27
200.81
27.44
33.54
29.44
87.8
Chl-a
(m)
89
0.51
4.20
0.98
1.21
0.70
57.9
SD depth
(NTU)
89
1.51
19.20
6.79
7.59
4.39
57.8
Turbidity
(g m-3)
89
1.19
15.00
6.80
7.30
3.26
44.6
TSS
(g m-3)
87
0.15
5.85
0.80
1.11
0.95
85.7
ISS
(g m-3)
87
0.81
12.80
6.00
6.26
2.97
47.5
OSS
(m-1)
80
0.0470
4.4867
0.4996
0.5915
0.5683
96.1
ap (675)
(m-1)
80
0.0031
0.2940
0.0595
0.0725
0.0554
76.4
aNAP (675)
(m-1)
80
0.0440
4.4051
0.4017
0.5190
0.5623
108.3
aφ (675)
80
0.0059
0.0285
0.0143
0.0156
0.0052
33.3
(m2 mg-1)
a*φ(675)
80
0.4553
1.4525
0.8946
0.8806
0.2468
28.0
aCDOM (440) (m-1)
80
0.0069
0.0333
0.0153
0.0158
0.0057
36.1
aCDOM (675) (m-1)
(nm-1)
80
0.0156
0.0185
0.0173
0.0172
0.0006
3.6
SCDOM
Chl-a – Fluorometrically measured chlorophyll-a concentration, SD depth – Secchi disk depth, TSS concentration of total suspended solids, ISS - concentration of inorganic suspended solids, OSS concentration of organic suspended solids, ap(675) - total particulate absorption coefficient at 675 nm,
aNAP(675) - absorption coefficient of non-algal particles at 675 nm, aφ(675) - absorption coefficient of
phytoplankton at 675 nm, a*φ(675) - chl-a specific absorption coefficient of phytoplankton at 675 nm,
aCDOM(440) and aCDOM(675) - absorption coefficients of CDOM at 440 and 675 nm, N - number of
samples, and CV - coefficient of variation.
Table 3.2 Descriptive statistics of the water quality parameters measured in 2009.
Chl-a
SD depth
Turbidity
TSS
ISS
OSS
ap (675)
aNAP (675)
aφ (675)
a*φ(675)
aCDOM (440)
aCDOM (675)
SCDOM
bbp(570)
bbp(660)
bbp(740)
(mg m-3)
(m)
(NTU)
(g m-3)
(g m-3)
(g m-3)
(m-1)
(m-1)
(m-1)
(m2 mg-1)
(m-1)
(m-1)
(nm-1)
(m-1)
(m-1)
(m-1)
N
Min
Max
Median
Mean
63
63
63
63
63
63
63
63
63
63
63
63
63
44
44
44
3.97
0.39
1.08
1.32
0.10
0.85
0.0650
0.0071
0.0411
0.0091
0.3530
0.0027
0.0165
0.0237
0.0155
0.0144
196.39
3.32
23.40
22.89
6.22
17.00
2.8214
0.1131
2.7204
0.0203
1.3489
0.0234
0.0210
0.1740
0.0944
0.0900
16.07
1.12
4.73
5.83
1.00
4.50
0.2922
0.0447
0.2282
0.0136
0.6475
0.0086
0.0182
0.0855
0.0562
0.0601
30.03
1.37
6.08
6.67
1.39
5.28
0.4764
0.0504
0.4261
0.0137
0.6852
0.0100
0.0182
0.0822
0.0526
0.0559
Standard
Deviation
CV
(%)
35.19
0.77
5.31
4.67
1.40
3.99
0.5371
0.0270
0.5220
0.0021
0.2203
0.0047
0.0009
0.0397
0.0248
0.0248
117.2
56.0
87.4
70.1
100.2
75.6
112.7
53.6
122.5
15.3
32.1
47.0
5.2
48.3
47.2
44.4
40
Secchi disk depth increased from an average of 1.18 m in midsummer to an
average of 1.95 m in late fall 2008 (Fig. 3.3). Spearman correlation coefficients, rs
indicated a statistically significant correlation (p < 0.01) of Secchi disk depth with chl-a
concentration (rs = -0.70), turbidity (rs = -0.87), and TSS (rs = -0.84), where Secchi disk
depth was highly correlated with OSS (rs = -0.88, p < 0.01) and only slightly correlated
with ISS (rs = -0.25, p = 0.02). Water transparency was minimally affected by the
presence of CDOM (rs = -0.12 for aCDOM(440), p = 0.27) in 2008. Comparable conditions
were found in 2009 (Fig. 3.3). Secchi disk depth decreased from an average of 1.27 m in
early spring to only 0.85 m in midsummer and showed a statistically significant
correlation (p<0.01) with chl-a concentration (rs = -0.90), turbidity (rs =-0.93), TSS (rs =0.94), OSS (rs =-0.96), and ISS (rs = -0.42). The water transparency was only slightly
affected by CDOM (rs =-0.25 for aCDOM(440), p = 0.04) in 2009.
5
2008
2009
SD depth, m
4
3
2
1
R² = 0.83
0
0
5
10
15
20
25
TSS, g m-3
Figure 3.3 Relationship of TSS concentration and Secchi disk depth for the Fremont
Lakes in 2008 and 2009.
41
TSS concentrations ranged from 1.2 to 15.0 g m-3 in 2008 and from 1.3 to 22.9 g
m-3 in 2009 and ISS concentrations ranged from 0.1 to 5.8 g m-3 in 2008 and from 0.1 to
6.2 g m-3 in 2009. The contribution of ISS to TSS ranged from 2.8 to 57.8% in 2008 and
from 2.4 to 61.1% in 2009 with a median contribution of 12.2% in 2008 and 19.2% in
2009. It was typically high for stations with low-to-moderate chl-a concentrations in
combination with moderate-to-high ISS concentrations. One of the highest contributions
of ISS to TSS was found in spring 2009 at a station with a chl-a concentration of 18.5 mg
m-3 and an ISS concentration of 4.9 g m-3. There were only four stations with a high
contribution of ISS to TSS in the presence of high chl-a concentrations (Fig. 3.4). These
stations correspond to the ones with chl-a concentrations from 129.5 to 200.8 mg m-3 in
fall 2008 and spring 2009 which coincided with relatively high ISS concentrations from
4.3 to 6.2 mg m-3.
1.00
2008
2009
ISS/TSS
0.75
FSL03
0.50
FSL03
Victory Lake
0.25
FSL03
0.00
0
50
100
Chl-a, mg m-3
150
200
Figure 3.4 Relationship of chl-a concentration and ISS/TSS. The four stations with a
high contribution of ISS to TSS in the presence of high chl-a concentrations are shown in
red.
42
The Fremont Lakes showed some of the typical optical properties of Case II
waters (Morel and Prieur, 1977; Gordon and Morel, 1983) and the seasonal differences in
the relationships of the water quality parameters may relativize the statistically significant
correlations (p < 0.01) found for chl-a and TSS concentrations in 2008 (rs = 0.64) and
2009 (rs = 0.90). In the first two weeks of July 2008, when the OSS concentrations reached
11.5 and 12.5 g m-3 and the chl-a concentrations were around 24 mg m-3, Fremont Lake 02
represented a typical Case II water without any correlation of chl-a concentration and
OSS (Fig. 3.5). The same is true for the last week of July when the chl-a concentration
reached its minimum of 15.5 mg m-3. In August, the chl-a concentration started to increase
while the OSS concentration remained remarkably constant around 5.5 g m-3. Only in
September and October, the chl-a concentration was correlated with the OSS concentration
and when the chl-a concentration reached its maximum of 47.0 mg m-3, the OSS
concentration increased to 11.0 g m-3. The ISS concentration averaged 1.5 g m-3 in Fremont
50
25
25
40
20
20
30
15
15
20
10
10
10
55
0
07/02 07/14 07/30 08/13 08/27 09/19 09/26 10/09 10/24 11/19
Chl-a
OSS
ISS
TSS and OSS, g m-3
Chl-a, mg m-3
Lake 02 and was independent of the chl-a concentration throughout summer and fall 2008.
00
Figure 3.5 Seasonal variations in the concentrations of chl-a, ISS, and OSS in Fremont
Lake 02 in summer and fall 2008.
43
3.2
Inherent bio-optical properties of the Fremont Lakes
The spectra of ap(λ), aNAP(λ), aφ(λ), and aCDOM(λ) showed a high variability in
2008 and 2009 (Fig. 3.6). The particulate absorption coefficient, ap(440), was typical for
inland and coastal waters and ranged from 0.292 to 6.860 m-1 in 2008 and from 0.287 to
7.482 m-1 in 2009 with coefficients of variation of 56.2% in 2008 and 87.7% in 2009. It
was highly correlated (p < 0.01) with the chl-a (rs = 0.82), TSS (rs = 0.78), OSS (rs =
0.75), and ISS (rs = 0.40) concentrations in 2008. The absorption coefficient of
phytoplankton, aφ(440), ranged from 0.061 to 5.598 m-1 and was highly correlated with
the chl-a concentration (rs = 0.80).
3
3
aNAP (λ), m-1
b 4
ap (λ), m-1
a 4
2
1
2
1
0
0
400 450 500 550 600 650 700 750
Wavelength, nm
400 450 500 550 600 650 700 750
Wavelength, nm
d 4
aCDOM (λ), m-1
aφ (λ), m-1
c 4
3
2
1
0
3
2
1
0
400 450 500 550 600 650 700 750
Wavelength, nm
400 450 500 550 600 650 700 750
Wavelength, nm
Figure 3.6 Representative spectra of the absorption coefficients of total particulates (a),
non-algal particles (b), phytoplankton (c), and CDOM (d) for 21 stations with chl-a
concentrations from 2.3 to 132.4 mg m-3.
44
These relationships were comparable to those in 2009, in which the correlations
of ap(440) with chl-a (rs = 0.92), TSS (rs = 0.90), and OSS (rs = 0.92) concentrations
were even stronger due to relatively low concentrations of ISS in spring 2009, and
aφ(440) was highly correlated with the chl-a concentration (rs = 0.96). Even though
ap(440) was related to the chl-a concentration in 2008 and 2009, it was still influenced by
aNAP(440). The contribution of aφ(440) to total particulate absorption ranged widely from
9.6 to 93.4% in 2008 and from 8.0 to 93.1% in 2009. Some of the spectra of aNAP(λ)
showed a deviation from the typical exponential shape (Fig. 3.6) comparable to spectra
taken in the Baltic Sea (Babin et al., 2003).
The contribution of phytoplankton pigments to total particulate absorption
increased substantially in the red spectral region and the contribution of aφ(675) to
ap(675) ranged from 47.9 to 99.1% in 2008 and from 56.2 to 98.8% in 2009.
Consequently, aφ(675) was highly correlated (p < 0.01) with the concentration of chl-a
(Fig. 3.7) in 2008 (rs = 0.87) and 2009 (rs = 0.98).
200
2008
2009
R² = 0.84
R² = 0.98
Chl-a, mg m-3
150
100
50
0
0
1
2
aφ (675),
3
4
5
m-1
Figure 3.7 Relationship of aφ(675) and chl-a concentration for the Fremont Lakes in
2008 and 2009.
45
The relatively small aNAP(675) and aCDOM(675) compared to aφ(675), in the waters
studied (Tab. 3.1 and 3.2), confirm the importance of the red spectral region for the
development of algorithms for the remote estimation of chl-a concentrations in inland
and coastal waters.
The spectra of the chl-a specific absorption coefficient of phytoplankton, a*φ(λ),
showed a relatively high variability and indicate seasonal differences in the physiological
status of the phytoplankton and species composition in the Fremont Lakes (Fig. 3.8a).
The phytoplankton consisted mainly of Bacillariophyta and Chlorophyta and the typical
spectral signature of Cryptophyta with absorption by chl-a, chl-c, carotenoids and the
distinctive absorption feature of phycoerythrin around 565 nm was only found in one lake
in fall 2008. The presence of cyanobacteria, indicated through the absorption feature of
phycocyanin around 620 nm, was typical in summer and fall 2008 and summer 2009
(Fig. 3.8b).
In 2008, the chl-a specific absorption coefficients of phytoplankton, a*φ(440) and
a*φ(675), were only slightly correlated (p < 0.05) with the chl-a concentration and
decreased with an increase in chl-a concentration (rs = -0.27 for a*φ(440) and rs = -0.39
for a*φ(675). In 2009, the relationship was less variable compared to 2008 with a slight
increase of a*φ(440) and a*φ(675) with an increase in chl-a concentration (rs = 0.68 for
a*φ(440), p < 0.05 and rs = 0.24 for a*φ(675), p = 0.06).
Even though a*φ(675) appears comparable in 2008 and 2009, a slightly negative
correlation of chl-a concentration and a*φ(675) was found for 2008, while no statistically
significant correlation was found for 2009 (Fig. 3.9). The differences of the relationships
aφ*(λ) versus chl-a concentration in 2008 and 2009 may be related to seasonal differences
46
in the species composition of the phytoplankton community and, therefore, differences in
the composition of phytoplankton pigments and pigment package effects (Bricaud et al.,
1995).
a*φ (λ), m2 mg-1
a 0.075
0.050
0.025
0.000
400
450
500
550
600
650
700
750
Wavelength, nm
b 1.25
Chl-a
a*φ (λ)/a*φ (440)
1.00
Chl-b, -c, and
β-carotene
0.75
Chl-a
0.50
PE
PC
0.25
0.00
400
450
500
550
600
650
Wavelength, nm
700
750
Figure 3.8 Representative spectra of the chl-a specific absorption coefficient of
phytoplankton (a) and normalized chl-a specific absorption coefficient of phytoplankton
(b) for the same stations shown in Fig. 3.6. The normalization of a*φ(λ) to a*φ(440)
enhances the spectral absorption features of different types of chlorophylls, carotenoids,
the cyanobacterial pigment phycocyanin (PC), and, for one lake in fall 2008,
phycoerythrin (PE).
a 0.08
b 0.04
a*φ (440), m2 mg-1
a*φ (675), m2 mg-1
47
0.06
0.04
0.02
0.00
2008
2009
0.03
0.02
0.01
0.00
0
50
100
150
Chl-a, mg m-3
200
0
50
100
150
Chl-a, mg m-3
200
Figure 3.9 Chl-a specific absorption coefficients of phytoplankton at 400 and 675 nm,
a*φ(440) and a*φ(675), shown versus chl-a concentration for the Fremont Lakes 2008 and
2009 datasets.
The absorption coefficients of CDOM, aCDOM(440) and aCDOM(675), ranged from
0.455 to 1.452 m-1 and 0.007 to 0.033 m-1 in 2008. The absorption budget at 440 nm
(Prieur and Sathyendranath, 1981) was typical for optically complex inland and coastal
waters with many samples found in the central part of the ternary plot (Fig. 3.10a). The
absorption at only one station, the station with the minimum chl-a and TSS
concentrations measured in 2008, was controlled by CDOM. This is a consequence of a
relatively high aCDOM(440) of 1.346 m-1 combined with an aφ(440) of only 0.165 m-1 and
an aNAP(440) of 0.127 m-1 for this station with a chl-a concentration of 2.3 mg m-3 and a
TSS concentration of 1.2 g m-3 at the time of field data collection in fall 2008. The effects
of the absorption by CDOM were still present for this station at 675 nm even though the
absorption budget at 675 nm was controlled by phytoplankton and NAP (Fig. 3.10b).
The absorption coefficients of CDOM, aCDOM(440) and aCDOM(675) ranged from
0.353 to 1.349 m-1 and 0.003 to 0.023 m-1 in 2009. These coefficients were statistically
significantly different (Mann-Whitney U test, p ≤ 0.01) from the coefficients measured in
48
2008. The absorption budgets at 440 and 675 nm showed less variability compared to
2008 and the samples were distributed in the ternary plots without any outliers in 2009
(Fig. 3.10c and d).
b
a
aφ(440)
aCDOM(440)
aNAP(440)
c
aCDOM(440)
aCDOM(675)
d
aφ(440)
aNAP(440)
aφ(675)
aCDOM(675)
aNAP(675)
aφ(675)
aNAP(675)
Figure 3.10 Ternary plots of the absorption budgets at 440 and 675 nm for 2008 (a, b)
and 2009 (c, d). The relative contribution of CDOM, phytoplankton, and NAP to
absorption was calculated through normalization of the individual absorption components
by x(x + y + z)-1, where x is the component of interest, and y and z designate the second
and third component. The distances from the apices of the triangle indicate the relative
contribution of the absorption components to absorption.
49
The spectral slope of the absorption coefficient of CDOM, SCDOM, ranged from
0.0156 to 0.0185 nm-1 in 2008 and from 0.0165 to 0.0210 nm-1 in 2009 and was typical
for inland (Binding et al., 2008) and coastal (Babin et al., 2003; Binding et al., 2005,
Kowalczuk et al., 2005) waters (Fig. 3.11). It was statistically significantly (p < 0.01)
correlated with aCDOM(440) in 2008 (rs = -0.36) and 2009 (rs = -0.60) and decreased with
an increase in aCDOM(440). Its variability seemed higher for smaller values of aCDOM(440)
comparable to the results from Babin (2003).
0.021
2008
2009
SCDOM, nm-1
0.020
0.019
0.018
0.017
0.016
R² = 0.33
0.015
0.0
0.5
1.0
aCDOM(440), m-1
1.5
2.0
Figure 3.11 Relationship of aCDOM(440) and SCDOM for the Fremont Lakes in 2008 and
2009.
The SCDOM values were normally distributed and averaged 0.0172 nm-1 in 2008
and 0.0182 nm-1 in 2009. We found less variability in SCDOM in the Fremont Lakes (Fig.
3.12) compared to the European coastal waters studied by Babin et al. (2003). Small
seasonal differences were found in 2008, where the average SCDOM was 0.0173 nm-1 in
the summer and decreased slightly to 0.0171 nm-1 in autumn, compared to 2009, where
the average SCDOM reached its maximum of 0.0185 nm-1 in spring and decreased to
0.0175 nm-1 in summer. The conspicuous intra-annual differences found in aCDOM (440)
50
and SCDOM (Fig. 3.11) may indicate differences in the molecular weight of humus
(Hayase and Tsubota, 1985; Carder et al., 1989) caused by seasonal differences in the
input and decomposition of organic materials and different sources of CDOM.
30
2008
2009
Frequency
25
20
Babin et al., 2003
15
10
5
0.01
0.01
0.01
0.01
0.01
0.01
0.02
0.02
0.02
0.02
0.02
0.02
0.02
0.02
0.02
0.02
0.02
0.02
0.02
0.02
0.02
0.02
0.02
0.02
0.02
0
0.012
0.014
0.016
0.018 0.020 0.022
SCDOM, nm-1
0.024
Figure 3.12 Frequency distributions of the slope of CDOM, SCDOM, measured in 2008
and 2009. The standard normal distribution curve for the European coastal waters studied
by Babin et al. (2003) for an average SCDOM of 0.0176 nm-1 and a standard deviation of
0.002 nm-1 is shown for comparison.
The spectra of the backscattering coefficients, bb(λ), were comparable to the
spectra of moderately turbid inland and coastal waters published by Kutser et al., 2009.
The backscattering coefficients ranged from 0.024 to 0.174 m-1 at 570 nm, from 0.016 to
0.095 m-1 at 660 nm and from 0.015 to 0.090 m-1 at 740 nm (Fig. 3.13). They decreased
from 660 to 740 nm for 10 out of 44 stations. This spectral dependence of bb(λ) is likely
to affect the accuracy of models for the remote estimation of chl-a concentrations, in
which a wavelength independent bb(λ) in the range from 660 through 750 nm is assumed,
even though the differences in bb(660) and bb(740) were not statistically significant
(Mann-Whitney U test, p = 0.48).
51
0.20
bb (λ), m-1
0.15
0.10
0.05
0.00
570
660
Wavelength
740
Figure 3.13 Spectra of the backscattering coefficient, bb(λ), collected in 2009. Only the
spectra without any saturation of the ECO Triplet sensor are shown. The sensor was
typically saturated in the NIR spectral region at a turbidity of 8 to 10 NTU and the
stations affected by the saturation of the sensor are expected to have a higher bb(λ).
The particulate backscattering coefficients, bbp(λ), were closely related to the OSS
concentrations and minimally related to the ISS concentrations (Fig. 3.14). The increase
in the correlation coefficient of the relationship bbp(λ) vs. TSS from rs = 0.80 for bbp(570)
to rs = 0.84 for bbp(740) confirms an increased effect of the particle concentration on
bbp(λ) at longer wavelengths.
The water quality parameters measured in 2008 and 2009 (Tables 3.1 and 3.2)
were statistically significantly different (Mann-Whitney U test, p ≤ 0.05), with the
exception of Secchi disk depth (p = 0.16), TSS (p = 0.06), and ISS (p = 0.60). The TSS
concentrations were not statistically significantly different in 2008 and 2009 (p = 0.06)
due to practically identical ISS concentrations in 2008 and 2009.
52
7.5
ISS
10.0
OSS
R² = 0.72
5.0
2.5
0.0
0.00
R² = 0.12
0.05
2.5
OSS, mg m-3
ISS, mg m-3
7.5
5.0
0.0
0.10
bbp (660), m-1
Figure 3.14 Relationship of the particulate backscattering coefficient, bbp(660), with ISS
and OSS.
3.3
Apparent bio-optical properties of the Fremont Lakes
The reflectance spectra collected in 2008 and 2009 (Fig. 3.15) showed a high
variability in the visible and NIR spectral regions and are comparable to spectra collected
around the world (Yang et al., 2009; Yacobi et al., 2011, Sun et al., 2011). In the blue
spectral region, the reflectance was low and without any distinctive spectral features due
to the combined absorption by phytoplankton pigments, NAP, and CDOM. The trough
related to the chl-a absorption maximum, typically found at 440 nm in oceanic waters
(Bidigare et al., 1990), was concealed by the absorption by NAP and CDOM for many
stations with chl-a concentrations up to 50 mg m-3. The reflectance increased in the green
region and reached a maximum around 570 nm, where the absorption by phytoplankton
pigments was minimal and reflectance was mainly affected by NAP and CDOM.
53
The reflectance in the red and NIR regions was characterized by a trough at 675 nm,
which corresponds to the red chl-a absorption maximum (Bidigare et al., 1990), and a peak
around 700 nm, related to a minimum in the combined absorption by phytoplankton
pigments and water (Vasilkov and Kopelevich, 1982; Gitelson, 1992 and 1993). The
magnitude of the reflectance was highly variable in the different spectral regions with
coefficients of variation from 42.1 to 81.9% in 2008 and 31.1 to 86.2% in 2009. The
highest coefficients of variation were found in the NIR in part due to very low reflectance
values (Fig. 3.15).
1.0
0.014
0.5
0.007
0.000
Coefficient of Variation
ρrs (λ), sr-1
a 0.021
0.0
400
500
600
700
800
900
Wavelength, nm
1.0
0.014
0.5
0.007
0.000
Coefficient of Variation
ρrs (λ), sr-1
b 0.021
0.0
400
500
600
700
800
900
Wavelength, nm
Figure 3.15 Remote sensing reflectance, ρrs(λ), spectra for the waters sampled in 2008
(a) and 2009 (b). The coefficient of variation of the reflectance is shown in grey.
54
The inverse reflectance relates directly to the combined absorption by
phytoplankton pigments, NAP, CDOM, and water (Eq. 1.1) and it showed several
distinctive spectral features at stations with high chl-a concentrations. It is relatively high
in the blue spectral region for the station shown in Fig. 3.16 due to the combined absorption
by phytoplankton pigments, NAP, and CDOM and the peak at 440 nm is not completely
concealed by the absorption by NAP and CDOM. The inverse reflectance spectrum for
this station shows a trough at 564 nm, where the absorption by pigments is minimal, and
a peak at 634 nm, possibly related to the presence of phycocyanin. The absorption peak
related to the chl-a absorption maximum in the red region is shifted slightly toward longer
wavelengths compared to the average peak position of 675 nm and the position of the trough
representative of the position of the minimal combined absorption by phytoplankton
pigments and water in the red region is found at a position of 705 nm for this station.
β-Carotene
750
Chlorophyll-a
Phycocyanin
ρrs-1, sr
500
Min 1
250
Min 2
Chlorophylls
CDOM
Tripton
0
400
450
500
550
600
650
Wavelength, nm
700
750
Figure 3.16 Inverse reflectance spectrum for a station with a chl-a concentration of 119.7
mg m-3. The positions of the maximal absorption by phytoplankton pigments, NAP, and
CDOM, the position of the minimal absorption by phytoplankton pigments (Min 1), and
the position of the minimal combined absorption by chl-a, NAP, and water (Min 2) are
indicated by arrows.
55
3.4
Bio-physical properties of Branched Oak Lake
The descriptive statistics of the water quality parameters measured in Branched
Oak Lake in 2011 (Tab. 3.3) indicate bio-physical and bio-optical conditions typical for
turbid highly productive inland waters (Sun et al., 2011) with chl-a concentrations from
70.8 to 159.7 mg m-3 and TSS concentrations from 11.2 to 50.4 g m-3. The chl-a
concentrations (Fig. 3.17) showed a normal distribution (Shapiro-Wilk test, p = 0.32) and
were significantly higher compared to the chl-a concentrations measured in the Fremont
Lakes in 2008 and 2009 (Mann-Whitney U test, p ≤ 0.01). ISS concentrations ranged
from 3.6 to 31.2 g m-3 in Branched Oak Lake and the composition of the suspended solids
was fundamentally different from the Fremont State Lakes. The contribution of ISS to
TSS ranged from 21.7 to 61.9% with a median contribution of 41.5% (Fig. 3.18). OSS
concentrations ranged from 7.5 to 19.2 g m-3.
40
Fremont Lakes
Branched Oak Lake
Frequency
30
20
10
0
010
60
50
110
160
100
150
Chl-a, mg m-3
210
200
250
Figure 3.17 Frequency distributions of the chl-a concentrations measured in 2008 and
2009 in the Fremont Lakes and in 2011 in Branched Oak Lake. The line represents the
standard normal distribution curve for Branched Oak Lake.
56
Table 3.3 Descriptive statistics of the water quality parameters measured in 2011.
N
Min
Max
Median
Mean
Standard
Deviation
CV
(%)
(mg m-3)
40
70.82
159.69
120.85
117.44
22.84
19.5
Chl-a
(m)
40
0.21
0.41
0.29
0.28
0.05
16.3
SD depth
(NTU)
40
11.78
35.97
19.58
20.48
6.36
31.1
Turbidity
(g m-3)
40
11.25
50.40
20.44
22.73
8.05
35.4
TSS
(g m-3)
40
3.60
31.20
8.00
9.77
5.95
60.9
ISS
(g m-3)
40
7.5
19.20
12.40
12.96
2.82
21.8
OSS
(m-1)
40
0.8778
2.0806
1.5178
1.5057
0.3006
20.0
ap (675)
(m-1)
40
0.0646
0.2418
0.1279
0.1345
0.0482
35.8
aNAP (675)
(m-1)
40
0.7529
1.9181
1.4144
1.3713
0.2875
21.0
aφ (675)
(m2 mg-1)
40
0.0072
0.0170
0.0118
0.0118
0.0018
15.3
a*φ(675)
40
0.5953
0.8862
0.7082
0.7175
0.0816
11.4
aCDOM (440) (m-1)
40
0.0071
0.0210
0.0117
0.0128
0.0040
31.3
aCDOM (675) (m-1)
(nm-1)
40
0.0157
0.0188
0.0175
0.0173
0.0009
5.2
SCDOM
Chl-a – Fluorometrically measured chlorophyll-a concentration, SD depth – Secchi disk depth, TSS concentration of total suspended solids, ISS - concentration of inorganic suspended solids, OSS concentration of organic suspended solids, ap(675) - total particulate absorption coefficient at 675 nm,
aNAP(675) - absorption coefficient of non-algal particles at 675 nm, aφ(675) - absorption coefficient of
phytoplankton at 675 nm, a*φ(675) - chl-a specific absorption coefficient of phytoplankton at 675 nm,
aCDOM(440) and aCDOM(675) - absorption coefficients of CDOM at 440 and 675 nm, N - number of
samples, and CV - coefficient of variation.
1.00
Fremont Lakes
Branched Oak Lake
ISS/TSS
0.75
0.50
0.25
0.00
0
50
100
Chl-a, mg m-3
150
200
Figure 3.18 Comparison of the relationship of chl-a concentration and ISS/TSS for the
Fremont Lakes and Branched Oak Lake.
57
3.5
Inherent and apparent bio-optical properties of Branched Oak
Lake
The spectra of ap(λ), aNAP(λ), aφ(λ), and aCDOM(λ) showed less variability
compared to the Fremont Lakes (Fig. 3.19). The particulate absorption coefficient,
ap(440), was typical for turbid highly productive inland waters. It ranged from 2.768 to
6.487 m-1 with a coefficient of variation of only 12.7% and seemed mainly affected by
the ISS concentration even though it was correlated (p < 0.05) with the chl-a (rs = 0.39),
TSS (rs = 0.67), OSS (rs = 0.45), and ISS (rs = 0.74) concentrations.
a 7.5
b 5.0
aNAP (λ), m-1
ap (λ), m-1
4.0
5.0
2.5
3.0
2.0
1.0
0.0
0.0
400 450 500 550 600 650 700 750
Wavelength, nm
400 450 500 550 600 650 700 750
Wavelength, nm
4.0
2.0
aCDOM (λ), m-1
c 2.5
aφ (λ), m-1
c 5.0
3.0
2.0
1.0
1.5
1.0
0.5
0.0
0.0
400 450 500 550 600 650 700 750
Wavelength, nm
400 450 500 550 600 650 700 750
Wavelength, nm
Figure 3.19 Representative spectra of the absorption coefficients of total particulates (a),
non-algal particles (b), phytoplankton (c), and CDOM (d) for 21 stations with chl-a
concentrations from 83.6 to 159.7 mg m-3.
58
The absorption coefficient of phytoplankton, aφ(440), ranged from 1.405 to 3.968
m-1 and was highly correlated (p < 0.01) with the chl-a concentration (rs = 0.62). There
was a considerable influence of aNAP(440) on ap(440) and the contribution of aφ(440) to
total particulate absorption ranged from 43.1 to 75.1% with a median contribution of 59.5%.
The contribution of absorption by phytoplankton to total particulate absorption increased
in the red spectral region, where the contribution of aφ(675) to ap(675) ranged from 85.2 to
95.5% and the median contribution was 90.8%. The absorption coefficient of phytoplankton,
aφ(675), was highly correlated (p < 0.01) with the chl-a concentration (rs = 0.72).
The spectra of the chl-a specific absorption coefficient of phytoplankton, a*φ(λ),
showed less variability compared to the Fremont Lakes 2008 and 2009 datasets. The
phytoplankton consisted mainly of Chlorophyta and Bacillariophyta and the typical
spectral signature of cyanobacteria with the absorption feature of phycocyanin around
620 nm was present throughout the summer (Fig. 3.20). Typical cyanobacteria included
Microcystis sp., Anabena sp., and Oscillatoria sp.. Some Euglenophyta were found at
stations with relatively high concentrations of TSS from 15.4 to 50.4 g m-3.
a*φ (λ), m2 mg-1
0.040
0.030
0.020
0.010
0.000
400
450
500
550
600
650
Wavelength, nm
700
750
Figure 3.20 Representative spectra of the chl-a specific absorption coefficient of
phytoplankton for the same stations shown in Fig. 3.20.
59
The chl-a specific absorption coefficients of phytoplankton, a*φ(440) and
a*φ(675), seemed consistent with the ones from the Fremont Lakes even though they were
not correlated with the chl-a concentration (rs = -0.18 for a*φ(440), p = 0.27 and rs = -
a 0.08
b 0.04
Fremont Lakes
a*φ (440), m2 mg-1
a*φ (675), m2 mg-1
0.19 for a*φ(675), p = 0.24) in Branched Oak Lake (Fig. 3.21).
Branched Oak Lake
0.06
0.04
0.02
0.00
0.03
0.02
0.01
0.00
0
50
100
150
-3
Chl-a, mg m
200
0
50
100
150
-3
Chl-a, mg m
200
Figure 3.21 Relationship of the chl-a specific absorption coefficients of phytoplankton at
400 and 675 nm, a*φ(440) and a*φ(675), and chl-a concentration.
The absorption coefficients of CDOM, aCDOM(440) and aCDOM(675), ranged from
0.595 to 0.886 m-1 and 0.007 to 0.021 m-1 and were comparable to the ones for the
Fremont Lakes. Even though there was substantial absorption by CDOM at 440 nm, this
was not reflected in the absorption budget at 440 nm, which was controlled by
phytoplankton and NAP in consequence of the high chl-a and TSS concentrations found
in Branched Oak Lake (Fig. 3.22a). The absorption budget at 675 nm was controlled by
phytoplankton, partially affected by NAP, and minimally affected by CDOM (Fig. 3.22b).
60
The spectral slope of the absorption coefficient of CDOM, SCDOM, ranged from
0.0157 to 0.0188 nm-1 and was practically identical with the spectral slope measured for
the Fremont Lakes (Mann-Whitney U test, p ≤ 0.10). It was statistically significantly
correlated (rs = -0.82, p < 0.01) with aCDOM(440) and decreased with an increase in
aCDOM(440).
a
b
aφ(440)
aφ(675)
Fremont Lakes
Branched Oak
Lake
aCDOM(440)
aNAP(440)
aCDOM(675)
aNAP(675)
Figure 3.22 Ternary plots of the absorption budgets at 440 and 675 nm for the Fremont
Lakes and Branched Oak Lake.
The reflectance spectra collected in 2011 represented the highly turbid conditions
in Branched Oak Lake (Le et al., 2009; Sun et al., 2011) and showed some variability in
the visible and NIR spectral regions (Fig. 3.23). In the blue spectral region, the
reflectance was even lower compared to the reflectance measured at the Fremont Lakes
and without any distinctive spectral features. The reflectance increased in the green
region and reached a maximum around 562 nm, a slightly shorter wavelength compared
to the Fremont Lakes, where the absorption by phytoplankton pigments was minimal.
The reflectance in the red and NIR regions was characterized by a trough around 620 nm,
which represents absorption by phycocyanin, and a trough at 675 nm, which is caused by
61
absorption by chl-a (Bidigare et al., 1990). The peak around 700 nm, related to the
minimum in the combined absorption by phytoplankton pigments and water (Vasilkov
and Kopelevich, 1982; Gitelson, 1992 and 1993), was slightly higher compared to the
peak in green region, which is typical in conditions with high chl-a and TSS
concentrations (Han, 1997). The magnitude of the reflectance was less variable compared
to the Fremont Lakes with coefficients of variation from 15.5 to 66.0%.
1.0
ρrs (λ), sr-1
0.8
0.014
0.6
0.4
0.007
0.2
0.000
Coefficient of Variation
0.021
0.0
400
500
600
700
Wavelength, nm
800
900
Figure 3.23 Remote sensing reflectance, ρrs(λ), spectra for the waters sampled in 2011.
The coefficient of variation of the reflectance is shown in grey.
62
Chapter 4. Return to a Simple Two-band NIR-red Model
for the Remote Estimation of Chl-a
in Turbid Productive Waters?
4.1
Importance of the red and NIR spectral regions for the
remote estimation of chl-a concentrations in inland and
coastal waters
Conventional maximum band ratio algorithms (O’Reilly et al., 2000) for the
remote estimation of chl-a concentrations in oceanic waters in form of the MERIS OC4E
and MODIS OC3M algorithms rely on band combinations in the blue and green spectral
regions. These spectral regions are highly affected by the absorption by NAP and CDOM
in inland and coastal waters and, thus, the algorithms developed for oceanic waters are
typically inapplicable in these waters.
The consequences of the application of the MERIS OC4E algorithm for the
remote estimation of chl-a concentrations in inland and coastal waters are exemplified by
spectra of aφ(λ), aNAP(λ), aCDOM(λ), and aw(λ) for a slightly turbid station in Fremont Lake
18 with a chl-a concentration of 4.6 mg m-3 and TSS concentration of 3.5 g m-3 (Fig. 4.1).
In the blue spectral region, the contribution of aφ(442.5) to total absorption was very
small with only 18.0% compared to contributions of 36.8% by aNAP(442.5) and 44.5% by
aCDOM(442.5). The contribution of aw(442.5) to total absorption was minimal. The
contributions of aφ(λ), aNAP(λ), and aCDOM(λ) to total absorption changed only slightly at
63
490 and 510 nm and the contribution of aw(λ) increased toward longer wavelengths. In
the green region, the contribution of aφ(560) to total absorption decreased to 6.2%, the
combined contribution by aNAP(560) and aCDOM(560) was 72.5%, and the contribution by
aw(560) increased to 21.3%.
a
4
- aw
- aCDOM
- aNAP
- aφ
a(λ), m-1
3
2
1
0
400
b
450
500
550
600
650
Wavelength, nm
a(λ), m-1
6
700
750
- aw
- aCDOM
- aNAP
- aφ
4
2
0
400
450
500
550
600
650
Wavelength, nm
700
750
Figure 4.1 Comparison of the absorption coefficients of phytoplankton, aφ(λ), non-algal
particles, aNAP(λ), CDOM, aCDOM(λ), and water, aw(λ), for two stations with chl-a
concentrations of 4.6 mg m-3 (a) and 58.1 mg m-3 (b). The chl-a absorption peak at 440
nm for the station with a chl-a concentration of 4.6 mg m-3 is completely concealed by
the absorption by NAP and CDOM (aw(λ) taken from Mueller, 2003).
64
The combined absorption coefficient of phytoplankton, NAP, CDOM, and water
was closely related to the absorption by NAP and CDOM for wavelengths from 400 to
565 nm and decreased exponentially towards longer wavelengths (R2 = 0.9995) from
2.1614 m-1 at 400 nm to 0.3198 m-1 at 565 nm.
Thus, in the waters studied, the MERIS OC4E algorithm, based on the maximal
ratio of ρ(442.5)/ρ(560), ρ(490)/ρ(560), and ρ(510)/ρ(560) and MODIS OC3M-551
algorithm, based on the maximal ratio of ρ(443)/ρ(551) and ρ(489)/ρ(551) were unable to
accurately estimate chl-a concentrations. The normalized absolute error (NAE) exceeded
186.6% when the MERIS OC4E algorithm was applied for the station in Fremont Lake
18 (Fig. 4.1) and the MNAE was still 34.4% when the same algorithm was applied for the
Fremont Lakes 2009 dataset with chl-a concentrations from 4.0 to 95.5 mg m-3 (Fig. 4.2).
Chl-aestimated OC4E, mg m-3
100
R² = 0.62
1:1 Line
75
50
25
0
0
25
50
75
Chl-ameasured, mg m-3
100
Figure 4.2. Application of the MERIS OC4E algorithm for the estimation of chl-a in
turbid inland waters.
65
In contrast to the blue and green spectral regions, the contributions of aNAP(675)
and aCDOM(675) to total absorption were only 2.0 and 4.5%, respectively, in the red
spectral region. The contribution of aφ(675) increased to 20.9% and the main contribution
came from aw(675) with 72.6%. The minimal absorption by NAP and CDOM and
distinctive absorption features of chl-a, with maximal absorption around 675 nm and
minimal combined absorption by water and phytoplankton pigments around 700 nm, in
addition to the well-known absorption coefficient of water (Kou et al., 1993; Pope and
Fry, 1997), made the red region suitable for the development of algorithms for the remote
estimation of chl-a concentrations in inland and coastal waters.
Many of these algorithms relied on the magnitude of the reflectance peak around
700 nm. With an increase in chl-a concentration, the magnitude of the absorption peak at
675 nm and its width increase and the point of the minimal combined absorption by
phytoplankton pigments and water, responsible for the reflectance peak around 700 nm,
shifts towards longer wavelengths (Vasilkov and Kopelevich, 1982; Gitelson, 1992). This
shift was observed in the Fremont Lakes where the point of minimal combined
absorption for seven representative stations shifted from 690 nm to 698 nm with an
increase in chl-a concentration from 2.3 to 132.4 mg m-3 (Fig. 4.3).
The position of the reflectance peak showed a close relationship with chl-a
concentration for chl-a concentrations from 4.0 to 200.8 mg m-3 for the Fremont Lakes
2008 and 2009 datasets (Fig. 4.4) even though the measurements of the reflectance peak
position were less reliable for some of the stations due to instrument noise when the
reflectance spectra were measured in very dark and cloudy conditions. This explains the
three outliers in the relationship of chl-a concentration and reflectance peak position (Fig.
66
4.4). These outliers represent stations with maximum reflectances of only 0.001 to 0.002
sr-1 in the red region for which the distinctive spectral features of the reflectance peak
were highly affected by instrument noise.
Sum of aw (λ) and aφ (λ), m-1
a 4.0
3.0
132.4 mg m-3
2.0
aw
1.0
2.3 mg m-3
0.0
400
450
500
550
600
650
Wavelength, nm
700
750
Sum of aw (λ) and aφ (λ), m-1
b 1.6
132.4 mg m-3
1.2
0.8
aw
2.3 mg m-3
0.4
680
690
700
Wavelength, nm
710
720
Figure 4.3 Spectra of the combined absorption coefficients of phytoplankton and water
for chl-a concentrations from 2.3 to 132.4 mg m-3 (a), and the position of the minimal
combined absorption by chl-a and water in the red region (b). The blue line represents
the absorption coefficient of water, aw(λ), (aw(λ) taken from Mueller (2003).
67
Peak Position, nm
705
700
695
690
685
R² = 0.78
680
0
50
100
Chl-a, mg m-3
150
200
Figure 4.4 Relationship of chl-a concentration and reflectance peak position for the
Fremont Lakes 2008 and 2009 datasets.
4.2
Empirical and semi-empirical NIR-red models for the remote
estimation of chl-a concentrations in inland and coastal waters
The characteristics of the reflectance peak in the red region were firstly applied
for the remote estimation of chl-a concentrations by Neville and Gower (1977). It was
considered a sun-induced chl-a fluorescence peak and an increase in reflectance at the
wavelength of 685 nm was found with an increase in chl-a concentration. In subsequent
studies, the peak magnitude at 685 nm above a baseline from 650 to 730 nm, the
fluorescence line height (FLH), was successfully related to the concentration of chl-a for
Saanich Inlet in British Columbia, Canada (Gower, 1980). When the FLH was applied for
the Fremont Lakes 2008 dataset, a very weak linear relationship of chl-a concentration
and FLH was found. The values of the peak magnitude at 685 nm were widely spread
from the regression line though (Fig. 4.5).
68
R² = 0.21
Fluorescence Line Height, sr-1
0.0010
0.0005
0.0000
0
25
50
75
100
-0.0005
-0.0010
Chl-a, mg m-3
Figure 4.5 Relationship of chl-a concentration and fluorescence line height for the
Fremont Lakes 2008 dataset.
Studies of the position of the reflectance peak around 700 nm revealed a shift of
the reflectance peak from 680 to 715 nm for different types of inland waters (Gitelson,
1992). Therefore, it was suggested (Gitelson, 1992) to use the reflectance line height
(RLH), the magnitude of the reflectance peak above a baseline from 675 to 730 nm,
instead of the FLH, for the estimation of chl-a concentrations in these waters (Fig. 4.6).
Even though the values were still widely scattered around the regression line, the
relationship of RLH versus chl-a concentration was reasonable (Fig. 4.7a), when the
RLH was applied to the Fremont Lakes 2008 dataset, and improved significantly (Fig.
4.7b), when the RLH was normalized to the reflectance minimum around 675 nm
(Gitelson 1992). The chl-a fluorescence peak at 685 nm and the reflectance peak around
700 nm are exploited in the fluorescence line height (FLH) and maximum chlorophyll
index (MCI) algorithms for the MERIS satellite sensor (Gower et al., 1999; Gower et al.,
2005) and a comparison of the MERIS FLH and MCI products for the Lake of the
69
Woods, USA/Canada, indicated a high potential of the MCI to successfully monitor
phytoplankton blooms in inland waters (Binding et al., 2011a; Binding et al., 2011b).
a 0.005
Spectrum
ρrs, sr-1
0.003
0.002
0.002
0.001
0.000
0.000
690
Wavelength, nm
c 0.0015
685 nm
660
690
680
700
720
Wavelength, nm
d 0.0015
740
RLH
698 nm
698 nm
0.0000
640
640
740
FLH
Spectrum
0.003
0.001
640
Baseline
0.004
740
RLH, sr-1
ρrs, sr-1
0.004
FLH, sr-1
b 0.005
Baseline
0.0010
0.0005
0.0000
-0.0015
640
690
Wavelength, nm
Wavelength, nm
740
Figure 4.6 Comparison of the fluorescence line height, FLH (a, c), and reflectance line
height, RLH (b, d), for a lake with a chl-a concentration of 18.0 mg m-3.
R² = 0.48
b 1.50
ρredmax/ρ675
Reflectance Line Height, sr-1
a 0.006
0.004
R² = 0.82
1.00
0.50
0.002
0.00
0.000
0
25
50
75
-3
Chl-a, mg m
100
0
25
50
75
Chl-a, mg m-3
Figure 4.7 Relationships of chl-a concentration with reflectance line height (a) and
normalized reflectance line height (b) for the Fremont Lakes 2008 dataset.
100
70
Simpler models, like the two band model ρ(705)/ρ(670), where ρ(705) is close to
the average position of the reflectance peak in the NIR region and ρ(670) is the
reflectance close to the chl-a absorption maximum, estimated chl-a concentrations very
accurately and the relationship of ρ(705)/ρ(670) vs. chl-a concentration was very close
(Fig. 4.8). This model (Eq. 1.7) was applied successfully in many studies since it was
devised in 1985 (Gitelson et al., 1985) and may provide an alternative to recently
developed more complex models (Gons, 1999, Gilerson et al., 2010, Yang et al., 2010).
R² = 0.91
2.5
ρ705 /ρ670
2.0
1.5
1.0
0.5
0.0
0
25
50
Chl-a, mg m-3
75
100
Figure 4.8 Relationship of ρ(705)/ρ(670) and chl-a concentration. The station marked in
red was excluded from the quadratic regression.
The optimal spectral regions for the three-band model (Eq. 1.5) and two-band
model (Eq. 1.7) were identified for the Fremont Lakes 2008 dataset through minimal
values of the standard error of the estimate (STE) for linear regressions between the
measured chl-a concentrations and model values (Dall’Olmo and Gitelson, 2005). The
initial spectral regions for the three-band model (Eq. 1.5) were selected identical to
Dall’Olmo and Gitelson (2005) with λ1 = 675 nm, λ2 = 710 nm, and λ3 = 750 nm.
71
To identify the optimal λ1, the chl-a concentration was regressed vs. the model in
the form [ρrs-1(λ1) – ρrs-1(710)] × ρrs(750) for ρrs-1(λ1) values from 400 to 900 nm. The
standard error of the estimate (STE) reached a minimum of 5.1 mg m-3 at a wavelength of
667 nm and λ1 was set to 667 nm. To estimate λ2, the chl-a concentration was regressed
vs. the model [ρrs-1(667) – ρrs-1(λ2)] × ρrs(750) and a minimum STE was found at 711 nm
with 4.8 mg m-3, relatively close to the wavelength selected by Dall’Olmo and Gitelson
(2005).
Finally, to derive λ3, the chl-a concentration was regressed vs. the model
[ρrs-1(667) – ρrs-1(711)] × ρrs(λ3). The optimal λ3 was found at a shorter wavelength
compared to Dall’Olmo et al. (2005), and, while the possible spectral region for λ3 was
relatively wide from around 710 nm to 755 nm, the STE increased slightly for
wavelengths of 730 nm and more. The minimum STE of 4.3 mg m-3 was found at a
wavelength of 724 nm and λ3 was set to 724 nm.
The iterative process was repeated to see if adjustments of the initially identified
wavelengths were required. To test the initially selected λ1, the chl-a concentration was
regressed vs. the model [ρrs-1(λ1) – ρrs-1(711)] × ρrs(724) and the optimal λ1 shifted
from 667 to 666 nm. Then the chl-a concentration was regressed vs. the model
[ρrs-1(667) – ρrs-1(λ2)] × ρrs(724) and a slight shift of λ2 from 711 to 712 nm was found.
The subsequent linear regression of the chl-a concentration vs. the model
[ρrs-1(667) – ρrs-1(712)] × ρrs(λ3) confirmed the previously identified λ3 at 724 nm with a
STE of 4.3 mg m-3 (Fig. 4.9).
72
a 20
Step 1 - λ1
b 20
667 nm
10
5
0
400
c 20
500
Step 3 - λ3
600
700
Wavelength, nm
800
400
d 20
724 nm
500
Step 4 - λ1
600
700
Wavelength, nm
800
666 nm
15
STE, mg m-3
STE, mg m-3
15
10
5
10
5
0
0
400
500
Step 5 - λ2
600
700
Wavelength, nm
800
400
f 20
712 nm
500
Step 6 - λ3
600
700
Wavelength, nm
800
724 nm
15
STE, mg m-3
15
STE, mg m-3
10
5
0
e 20
711 nm
15
STE, mg m-3
STE, mg m-3
15
Step 2 - λ2
10
5
10
5
0
0
400
500
600
700
Wavelength, nm
800
400
500
600
700
Wavelength, nm
800
Figure 4.9 Identification of the optimal spectral regions for the estimation of chl-a with
the three-band model (Eq. 1.5). The identification of λ1, λ2, and λ3 for the three-band
model involved the reduction of the standard error of the estimate, STE, in an iterative
process. The vertical lines represent the spectral regions with the smallest possible STE.
73
The initial spectral regions for the two-band model (Eq. 1.7) were selected
identical to Gitelson et al. (1985) with λ1 = 670 nm and λ2 = 705 nm and the spectral
regions with the smallest possible STE were identified through linear regressions of the
chl-a concentrations vs. the model in the forms ρrs-1(λ1) × ρrs(705) and ρrs-1(664) × ρrs(λ2).
The finally selected optimal wavelengths for the two-band model (Eq. 1.7) were only
slightly different from the optimal λ1 and λ2 for the three-band model (Eq. 1.5) with λ1 =
666 nm and λ2 = 713 nm and a minimum STE of 4.5 mg-3 (Fig. 4.10).
a 20
Step 1 - λ1
b 20
664 nm
10
5
10
5
0
0
400
c 20
500
Step 3 - λ1
600
700
Wavelength, nm
800
400
b 20
666 nm
500
Step 4 - λ2
600
700
Wavelength, nm
800
713 nm
15
STE, mg m-3
15
STE, mg m-3
713 nm
15
STE, mg m-3
STE, mg m-3
15
Step 2 - λ2
10
5
10
5
0
0
400
500
600
700
Wavelength, nm
800
400
500
600
700
Wavelength, nm
800
Figure 4.10 Identification of the optimal spectral regions for the estimation of chl-a with
the two-band model (Eq. 1.7).
74
4.3
Calibration of the NIR-red models for the remote estimation
of low-to-moderate chl-a concentrations
Relationships between the fluorometrically measured chl-a concentrations and
model values were established for the dataset taken in 2008 with chl-a concentrations
from 2.3 to 81.2 mg m-3. The three-band model (Eq. 1.5) and two-band model (Eq. 1.7)
were calibrated for the spectral bands of MERIS, and the two-band model (Eq. 1.6) was
calibrated for the spectral bands of MODIS. The optimal band position for λ1
corresponded closely to MERIS band 7 (660 – 670 nm) and MODIS band 13 (662 – 672
nm) and the position for λ2 for the three-band model (Eq. 1.5) and two-band model (Eq.
1.7) was similar to MERIS band 9 (703.75 – 713.75 nm), whereas MODIS lacked a band
in the relevant spectral region. The position of the optimal λ3 for the three-band model
(Eq. 1.5) and special case two-band model (Eq. 1.6) at 724 nm coincided with one of the
atmospheric absorption bands of water vapor (Grossmann and Browell, 1989), and was,
therefore, inapplicable for the estimation of chl-a concentrations with satellite sensors.
The STE for the identification of the optimal λ3 increased only slightly from 725 to 755
nm and λ3 was set to MERIS band 10 (750.0 - 757.5 nm) for the three-band model (Eq.
1.5) and MODIS band 15 (743 - 753 nm) for the two-band model (Eq. 1.6):
MERIS three-band model = [ρrs-1(665) – ρrs-1(709)] × ρrs(754)
(4.1)
MERIS two-band model = ρrs-1(665) × ρrs(709)
(4.2)
where ρ-1(665) is the simulated inverse reflectance for MERIS band 7 (centered at
665.00 nm), ρ-1(709) is the simulated inverse reflectance for MERIS band 9 (centered at
75
708.75 nm), and ρ(754) is the simulated reflectance for MERIS band 10 (centered 753.75
nm).
MODIS two-band model = ρrs-1(667) × ρrs(748)
(4.3)
where ρ-1(667) is the simulated inverse reflectance for MODIS band 13 (centered at 667
nm) and ρ(748) is the simulated reflectance for MODIS band 15 (centered at 748 nm).
For chl-a concentrations from 2.3 to 81.2 mg m-3, the MERIS three-band model
(Eq. 4.1) and MERIS two-band model (Eq. 4.2) showed very close relationships (R2 =
0.95) with chl-a concentration, which differed only slightly from linear relationships (Fig.
4.11a and 4.11b). There was a reasonable linear relationship (R2 = 0.75) of chl-a
concentrations and the MODIS two-band model (Eq. 4.3), though the spread of the points
from the best fit function is apparent for stations with chl-a concentrations up to 25 mg
m-3 (R2 = 0.04), where the reliability of this model was very low (Fig. 4.11c). The
algorithms for the remote estimation of chl-a concentrations, identified through the
calibration of the models, were established in the form:
Chl-a = 315.50 × (MERIS 3BM)2 + 215.95 × (MERIS 3BM) + 25.66
(4.4)
for the MERIS three-band model (Eq. 4.1)
Chl-a = 25.28 × (MERIS 2BM)2 + 14.85 × (MERIS 2BM) – 15.18
(4.5)
for the MERIS two-band model (Eq. 4.2) and
Chl-a = 190.34 × (MODIS 2BM) – 32.45
for the MODIS two-band model (Eq. 4.3).
(4.6)
76
3-band Model - MERIS
a 0.25
R² = 0.95
0.00
0
25
50
75
100
-0.25
Chl-a, mg m-3
2-band Model - MERIS
b 2.00
R² = 0.95
1.50
1.00
0.50
0.00
0
25
50
Chl-a, mg
75
c 0.75
2-band Model - MODIS
100
m-3
R² = 0.75
0.50
0.25
0.00
0
25
50
75
100
Chl-a, mg m-3
Figure 4.11 Calibration of the Eq. 4.1 MERIS three-band model (a), Eq. 4.2 MERIS twoband model (b), and the Eq. 4.3 MODIS two-band model (c), for the estimation of chl-a
concentrations up to 85 mg m-3.
77
4.4
Validation of the NIR-red models for the estimation of lowto-moderate chl-a concentrations
The independent and statistically significantly different dataset taken in 2009 was
used for the validation of the algorithms. Firstly, the reflectances measured in 2009 were
simulated in the spectral bands of MERIS and MODIS and the model values were
calculated by Eqs. 4.1 – 4.3. Secondly, the chl-a concentrations were estimated by the
algorithms shown in Eqs. 4.4 – 4.6 and, finally, the estimated chl-a concentrations,
chl-aestimated, were compared with the observed chl-a concentrations, chl-aobserved.
4.4.1 MERIS three-band model and MERIS two-band model
For chl-a concentrations from 4.0 to 95.5 mg m-3, the results of the validation
(Figs. 4.12 and 4.13, Tab. 4.1) were practically identical for the MERIS three-band and
two-band algorithms (Eqs. 4.4 and 4.5). The mean absolute error, MAE, for the
estimation of chl-a concentrations from 4.0 to 95.5 mg m-3 was 2.5 mg m-3 for the
MERIS three-band algorithm and 2.3 mg m-3 for the MERIS two-band algorithm. This
small difference in the MAE was not statistically significant (Student’s t-test, p = 0.63).
For chl-a concentrations in the range from 4.0 to 24.2 mg m-3, the difference in the
MAE was statistically significant (Student’s t-test, p < 0.01) and the MERIS two-band
algorithm was more accurate than the MERIS three-band algorithm, with an MAE of 1.2
mg m-3 compared to 1.9 mg m-3 for the MERIS three-band algorithm. The mean normalized
absolute error, MNAE, for the MERIS two-band algorithm was 11.5% compared to an
MNAE of 21.5% for the MERIS three-band algorithm (Figs. 4.14 and 4.15, Tab. 4.2).
78
Chl-aestimated, mg m-3
100
3-Band Model - MERIS
75
50
25
MAE = 2.5 mg m-3
RMSE = 3.3 mg m-3
MNAE = 18.0 %
0
0
25
50
Chl-aobserved, mg
75
100
m-3
Figure 4.12 Application of the MERIS three-band model, calibrated from the Fremont
Lakes 2008 dataset, for the estimation of chl-a concentrations up to 100 mg m-3 for the
Fremont Lakes 2009 dataset.
Chl-aestimated, mg m-3
100
2-Band Model - MERIS
75
50
25
MAE = 2.3 mg m-3
RMSE = 3.6 mg m-3
MNAE = 11.6 %
0
0
25
50
Chl-aobserved, mg
75
100
m-3
Figure 4.13 Application of the MERIS two-band model, calibrated from the Fremont
Lakes 2008 dataset, for the estimation of chl-a concentrations up to 100 mg m-3 for the
Fremont Lakes 2009 dataset.
79
Chl-aestimated, mg m-3
25
3-Band Model - MERIS
20
15
10
MAE = 1.9 mg m-3
RMSE = 2.1 mg m-3
MNAE = 21.5 %
5
0
0
5
10
15
Chl-aobserved, mg
20
25
m-3
Figure 4.14 Application of the MERIS three-band model, calibrated from the Fremont
Lakes 2008 dataset, for the estimation of chl-a concentrations up to 25 mg m-3 for the
Fremont Lakes 2009 dataset.
Chl-aestimated, mg m-3
25
2-Band Model - MERIS
20
15
10
MAE = 1.2 mg m-3
RMSE = 1.4 mg m-3
MNAE = 11.5 %
5
0
0
5
10
15
Chl-aobserved, mg
20
25
m-3
Figure 4.15 Application of the MERIS two-band model, calibrated from the Fremont
Lakes 2008 dataset, for the estimation of chl-a concentrations up to 25 mg m-3 for the
Fremont Lakes 2009 dataset.
80
The difference in the performance of these algorithms for the estimation of chl-a
concentrations from 4.0 to 24.2 mg m-3 is related to the use of λ3 and probably to the
different relationships of the TSS concentration and the reflectances at λ1 and λ2
compared to λ3 (Fig. 4.16). The reflectance at λ3 is practically insensitive to absorption by
pigments, NAP, and CDOM, but is susceptible to scattering by suspended particles and
is, consequently, highly correlated with TSS (rs = 0.65, p < 0.01) and the particulate
backscattering coefficient at λ3 (rs = 0.87 for ρrs(754) vs. bbp(740), p < 0.01).
For chl-a concentrations from 4.0 to 24.2 mg m-3, the correlation between bbp(λ)
and TSS concentrations increased consistently towards longer wavelengths – from rs =
0.69 at 570 nm to rs = 0.72 at 660 nm and to rs = 0.75 at 740. This indicates an increased
effect of the particle concentration on bb (λ) with an increase in wavelength and, thereby,
invalidates the assumption of spectral independence of bb(λ) in the wavelength range
from λ1 through λ3 for the three-band model (Dall’Olmo et al., 2003).
665 nm y = 0.00030x + 0.00089
709 nm y = 0.00033x + 0.00037
754 nm y = 0.00008x + 0.00012
ρrs(λ), sr-1
0.0050
665 nm
709 nm
0.0025
754 nm
0.0000
0.0
2.5
5.0
TSS, mg L-1
7.5
10.0
Figure 4.16 Relationship of TSS concentration and remote sensing reflectance, Rrs(λ), at
665, 709, and 754 nm for stations with chl-a concentrations from 4.0 to 24.2 mg m-3 in
2009.
81
Thus, the effects of bb(λ) might not be fully removed if ρ(λ3) is used and could
introduce uncertainties in the estimation of chl-a concentrations if the MERIS three band
model is applied. This is especially pronounced for chl-a concentrations from 4.0 to 24.2
mg m-3, where the reflectance is highly affected by scattering by suspended particles and
differences in the sensitivity of bb(λ) to suspended particles could cause significant
changes in the output of the MERIS three-band model. Therefore, even though the
algorithm was established for a wide range of bio-physical and bio-optical water quality
parameters, with ISS concentrations from 0.1 to 5.8 g m-3 and OSS concentrations from
0.8 to 12.8 g m-3, some caution is advised if this algorithm is applied to waters with
different ranges of ISS and OSS concentrations.
4.4.2 Comparison of the enhanced three-band index with the MERIS threeband model and MERIS two-band model
Recently, Yang et al. (2010) modified the three-band model and introduced the
enhanced three-band index in the form:
[Chl-a] ∝ [ρrs-1(λ1) - ρrs-1(λ2)] × [ρrs-1(λ3) - ρrs-1(λ2)]-1
(4.7)
where the term ρrs(λ3) was replaced by [ρrs-1(λ3) – ρrs-1(λ2)]-1 to compensate for absorption
and scattering by suspended particles at λ3 in highly turbid waters. We validated the
relationship of chl-a concentrations and the enhanced three-band index, established with
simulated MERIS spectral bands for chl-a concentrations from 30 to 120 mg m-3 at Lake
Kasumigure, Japan (Yang et al., 2010), for the Fremont Lakes 2009 dataset.
82
Its accuracy was comparable to the MERIS three-band algorithm with a MAE of
3.0 mg m-3 for chl-a concentrations from 4.0 to 95.5 mg m-3 and a MAE of 2.5 mg m-3 for
chl-a concentrations from 4.0 to 24.2 mg m-3 (Fig. 4.17), and it was much less accurate
than the MERIS two-band algorithm for the estimation of chl-a concentrations from 4.0
to 24.2 mg m-3 (Tabs. 4.1 and 4.2).
Chl-aestimated, mg m-3
a 100
Enhanced 3-Band Index
75
50
25
MAE = 3.0 mg m-3
RMSE = 4.0 mg m-3
MNAE = 18.5 %
0
0
25
50
75
Chl-aobserved, mg
Chl-aestimated, mg m-3
b 25
100
m-3
Enhanced 3-Band Index
20
15
10
MAE = 2.5 mg m-3
RMSE = 2.9 mg m-3
MNAE = 22.4 %
5
0
0
5
10
15
Chl-aobserved, mg
20
25
m-3
Figure 4.17 Application of the enhanced three-band index, calibrated for Lake
Kasumigure, Japan (Yang et al., 2010), for the estimation of chl-a concentrations up to
100 mg m-3 (a) and up to 25 mg m-3 (b) for the Fremont Lakes 2009 dataset.
83
4.4.3 MODIS two-band model
The MODIS two-band algorithm (Eq. 4.6) was less accurate than the MERIS
three-band algorithm (Eq. 4.4), the MERIS two-band algorithm (Eq. 4.5), and the
enhanced three-band index (Eq. 4.7) for the estimation of chl-a concentrations (Fig 4.18).
This was especially pronounced for chl-a concentrations up to 25 mg m-3, where the
MAE was 3.1 mg m-3 and the MNAE was 30.9% (Tabs. 4.1 and 4.2).
Chl-aestimated, mg m-3
a 100
2-Band Model - MODIS
75
50
25
MAE = 4.6 mg m-3
RMSE = 6.1 mg m-3
MNAE = 27.6 %
0
0
25
50
75
Chl-aobserved, mg
Chl-aestimated, mg m-3
b 25
100
m-3
2-Band Model - MODIS
20
15
10
MAE = 3.1 mg m-3
RMSE = 3.8 mg m-3
MNAE = 30.9 %
5
0
0
5
10
15
20
25
Chl-aobserved, mg m-3
Figure 4.18 Application of the MODIS two-band model, calibrated from the Fremont
Lakes 2008 dataset, for the estimation of chl-a concentrations up to 100 mg m-3 (a) and up
to 25 mg m-3 (b) for the Fremont Lakes 2009 dataset.
84
These results are expected. Stumpf and Tyler (1988) were well aware of the
limitations related to the use of spectral bands in the red and NIR regions and suggested
to apply the model for the estimation of moderate to high chl-a concentrations typical for
phytoplankton blooms.
4.4.4 Comparison of Gons’ NIR-red model and the MERIS three-band and
two-band models
We compared the performance of Gons’ algorithm (Eq. 4.8) with simulated bands
7, 9 and 12 of the MERIS level-2 standard product (Gons et al., 2005):
[Chl - a] =
ρ (709)
1.060
(a w (709) + bb ) - a w (665) - bb
ρ (665)
0.0161
(4.8)
where ρ(665) is the reflectance for MERIS band 7 (centered at 665.00 nm), ρ(709) is the
reflectance for MERIS band 9 (centered at 708.75 nm), aw(665) and aw(709) were set to
0.4 and 0.7, bb was derived from MERIS band 12 (centered at 778.75 nm), p was set to
1.06, and a*φ(665) was 0.0161, to the performance of the MERIS three-band algorithm
(Eq. 4.1) and MERIS two-band algorithm (Eq. 4.2). This model is only slightly different
from the MERIS two-band model (Eq. 4.2) and if the MERIS two-band model is
reformulated in terms of a(λ) and bb(λ), it is possible to derive Gons’ model, assuming
the effects of aNAP(λ) and aCDOM(λ) are insignificant compared to the effects of aφ(λ) and
aw(λ) in the relevant spectral regions, and bb is wavelength independent.
85
We found a slightly non-linear relationship (Fig. 4.19a) between estimated and
measured chl-a concentrations and the applied algorithm (Eq. 4.8) significantly
underestimated chl-a concentrations with an MAE of 5.9 mg m-3 for chl-a concentrations
from 4.0 to 95.5 mg m-3 and an MAE of 3.1 mg m-3 for chl-a concentrations from 4.0 to
24.2 mg m-3 (Tabs. 4.1 and 4.2). We, therefore, re-parameterized the algorithm for the
Fremont Lakes 2008 dataset (Eq. 4.9). The coefficients for a*φ(665) and p were found
through the minimal STE. The identified a*φ(665) of 0.0115 m-1 differed only slightly
from the median a*φ(665) of 0.0112 m-1 we measured in the laboratory for the Fremont
Lakes 2008 dataset. The exponent p was set to 1.024. The re-parameterized algorithm had
the form:
[Chl - a] =
ρ (709)
1.024
(a w (709) + bb ) - a w (665) - bb
ρ (665)
0.0115
(4.9)
where ρ(665) is the reflectance for MERIS band 7 (centered at 665.00 nm), ρ(709) is the
reflectance for MERIS band 9 (centered at 708.75 nm), aw(665) and aw(709) were set to
0.4 and 0.7, bb was derived from MERIS band 12 (centered at 778.75 nm), p was set to
1.024, and a*φ(665) was 0.0115.
When we applied the re-parameterized algorithm (Eq. 4.9) for the Fremont Lakes
2009 dataset (Fig. 4.19b), the MAE was 2.3 mg m-3 for chl-a concentrations from 4.0 to
95.5 mg m-3 and 1.4 mg m-3 for chl-a concentrations from 4.0 to 24.2 mg m-3 and the
results were practically identical to the results for the MERIS two-band algorithm (Tabs.
4.1 and 4.2).
86
Thus, the difference between the results from Gons’ algorithm and the MERIS
two-band algorithm seems mainly caused by the difference in a*φ(665), whereas the reparameterization of p seemed to have only a limited effect on the relationship between
the measured and estimated chl-a concentrations. The model is sensitive to a*φ(665) and
the difference in a*φ(665) resulted in a considerable change in the slope of the
relationship between the estimated and measured chl-a concentrations.
Chl-aestimated, mg m-3
a 100
Gons' Model - aφ*(665) = 0.0161
75
50
25
MAE = 5.9 mg m-3
RMSE = 8.6 mg m-3
MNAE = 26.7 %
0
0
25
50
Chl-aobserved, mg
b 100
Chl-aestimated, mg m-3
1:1 Line
75
100
m-3
Gons' Model - aφ*(665) = 0.0115
75
50
25
MAE = 2.3 mg m-3
RMSE = 3.5 mg m-3
MNAE = 12.7 %
0
0
25
50
75
100
Chl-aobserved, mg m-3
Figure 4.19 Comparison of Gons’ original algorithm (a), (Eq. 4.8), and re-parameterized
algorithm (b), (Eq. 4.9), for the estimation of chl-a concentrations up to 100 mg m-3 for
the Fremont Lakes 2009 dataset.
87
The model was only slightly affected by variations in bb and 99.7% of the
variability in the model (Fig. 4.20) were explained by ρ(665)-1 × ρ(709). These findings
indicate a high potential for a simpler two-band NIR-red algorithm, without the retrieval
of bb from the NIR wavelengths, to accurately estimate chl-a concentrations in turbid
productive waters.
Chl-aGons' model, mg m-3
100
Gons' Model - aφ*(665) = 0.0115
75
50
25
R² = 0.997
0
0.6
0.8
1.0
1.2
1.4
1.6
1.8
ρ(665)-1 × ρ(709)
Figure 4.20 Comparison of ρ(665)-1 × ρ(709) and the chl-a concentrations estimated
through Gons’ re-parameterized algorithm (Eq. 4.9).
The MERIS three-band model (Eq. 4.1) and MERIS two-band model (Eq. 4.2)
might have the same sensitivity to a*φ(665). Even though a*φ(λ) was statistically
significantly different for the Fremont Lakes 2008 calibration dataset and Fremont Lakes
2009 validation dataset, it is possible that this difference was sufficiently small to
minimally affect the results of the chl-a estimation. If the MERIS three-band algorithm
(Eq. 4.4) and MERIS two-band algorithm (Eq. 4.5) were applied to waters with a very
different a*φ(λ), these differences might affect the results of the chl-a estimation with
88
these algorithms considerably and future studies should consider the investigation of the
effects of a*φ(λ) on the estimation of chl-a concentrations with these red-NIR algorithms.
4.4.5 Advanced MERIS three-band model and Advanced MERIS two-band
model
In a slightly different approach, Gilerson et al. (2010) studied comprehensive
synthetic datasets of reflectance spectra and inherent optical properties, combined with
datasets from inland and coastal waters, to evaluate the sensitivities of the MERIS threeband and two-band models (Eqs. 4.1 and 4.2) and the MODIS two-band model (Eq. 4.3)
to changes in the optical water quality parameters. In contrast to the MERIS three-band
and two-band models, which were mainly controlled by terms related to aw(λ) and a*φ(λ)
and relatively insensitive to aCDOM(λ) and bb(λ), it was shown that the MODIS two-band
model was very sensitive to changes in the optical water quality parameters and it was
found inapplicable for the estimation of chl-a concentrations up to 20 mg m-3.
Gilerson et al. (2010) reformulated the MERIS three-band and two-band models
in terms of a(λ) and bb(λ) and separated a(λ) into aφ(λ), aNAP(λ), aCDOM(λ), and aw(λ), to
develop simplified semi-empirical models for the remote estimation of chl-a
concentrations, which were slightly different from Gons’ model (Eqs. 1.3 and 1.4):
MERIS three-band model
Chl - a =
[3BM( a w (754)) - a w (708) + a w (665)]
*
a φ (665)
(4.10)
89
MERIS two-band model
Chl - a =
[ 2BM ( a w (708)) - a w (665)]
*
a φ (665)
(4.11)
where 3BM and 2BM represent the model values derived from Eqs. 4.1 and 4.2,
respectively, ρ(665) is the reflectance for MERIS band 7 (centered at 665.00 nm), ρ(709)
is the reflectance for MERIS band 9 (centered at 708.75 nm), and ρ(754) is the
reflectance for MERIS band 10 (centered at 753.75 nm).
The analysis of the synthetic datasets (Gilerson et al. 2010) showed a high
sensitivity of the MERIS three-band and two-band models (Eqs. 4.1 and 4.2) to the
magnitude and spectral shape of a*φ(λ). The typical ranges of a*φ(675) and a*φ(665) and
their dependence on chl-a concentration were studied for several datasets collected at the
Fremont Lakes in 2008 and 2009, in Chesapeake Bay in 2005, and in the Long Island
Sound from 2006 to 2008. For these datasets, the mean values and the variability of
a*φ(675) decreased with an increase in chl-a concentration due to pigment package
effects (Bricaud et al., 1995). The substitution of a*φ(665) with a value derived from
these datasets and of aw(λ) with values from the literature resulted in the final algorithms:
Chl - a = (113 .36 × R3 + 16.45)1.124
(4.12)
for the advanced MERIS three-band model and
Chl - a = (35.75 × R2 - 19.30)1.124
for the advanced MERIS two-band model.
(4.13)
90
These algorithms (Eqs. 4.12 and 4.13) were applied to estimate chl-a
concentrations for the Fremont Lakes 2009 dataset. The advanced MERIS three-band
algorithm estimated chl-a concentrations in the Fremont Lakes accurately with an MAE
of 3.4 mg m-3 and an MNAE of 15.6% for chl-a concentrations from 4.0 to 95.5 mg m-3
(Tab. 4.1). The RMSE was 5.8 mg m-3. This difference in the MAE compared to the
RMSE is related to problems in the estimation of high chl-a concentrations (Fig. 4.21).
The algorithm worked very accurately for chl-a concentrations from 4.0 to 24.2 mg m-3
with an MAE of 1.4 mg m-3 and an MNAE of 15.2% (Tab. 4.2) and the results for the
estimation of chl-a concentrations up to 25 mg m-3 were only slightly different from those
for the MERIS two-band algorithm and advanced MERIS two-band algorithm (Tabs. 4.1
and 4.2).
Chl-aestimated, mg m-3
100
Advanced MERIS Three-band Model 1:1 Line
75
50
MAE = 3.4 mg m-3
RMSE = 5.8 mg m-3
MNAE = 15.6 %
25
0
0
25
50
Chl-aobserved, mg
75
100
m-3
Figure 4.21 Application of the advanced MERIS three-band model (Gilerson et al.,
2010) for the estimation of chl-a concentrations up to 100 mg m-3 for the Fremont Lakes
2009 dataset.
91
4.5
Comparison of the performances of the NIR-red algorithms
To compare the accuracy of the algorithms, an ANVOA, followed by Tukey’s
HSD multiple comparison procedure, was performed. Tukey’s HSD indicated statistically
significant differences, when the absolute estimation errors from the MERIS two-band
algorithm (Eq. 4.5), the algorithm with the smallest MAE for chl-a concentrations from
4.0 to 95.5 mg m-3 and 4.0 to 24.2 mg m-3, were compared to the errors of the MODIS
two-band algorithm (Eq. 4.6) and Gons’ algorithm (Eq. 4.8) for chl-a concentrations from
4.0 to 95.5 mg m-3 (p < 0.05).
Differences were more pronounced for chl-a concentrations from 4.0 to 24.2 mg
m-3, where Tukey’s HSD indicated statistically significant differences (p < 0.05), when
the errors of the MERIS two-band algorithm (Eq. 4.5) were compared to the errors of the
MERIS three-band algorithm (Eq. 4.4), the MODIS two-band algorithm (Eq. 4.6), the
MERIS enhanced three-band index (Eq. 4.7), and Gon’s algorithm (Eq. 4.8). The MERIS
two-band algorithm (Eq. 4.5), the re-parameterized Gons’ algorithm (Eq. 4.9), the
advanced MERIS three-band algorithm (Eq. 4.10), and the advanced MERIS two-band
algorithm (Eq. 4.11) successfully estimated chl-a concentrations up to 25 mg m-3 with
MAEs from 1.2 to 1.4 mg m-3 and MNAEs from 11.5 to 15.2% (Tab. 4.2).
These results indicate a high potential of the simple MERIS two-band algorithm
for the accurate estimation of chl-a concentrations in inland and coastal waters with chl-a
concentrations up to 25 mg m-3, without any reduction in accuracy compared to the more
complex algorithms, even though more research is required to analyze the sensitivity of
the developed algorithm to differences in a*φ(665). Interestingly enough, Yacobi et al.
92
(2011), studying performances of the three- and two-band models using reflectance
spectra taken at Lake Kinneret, came to the same conclusion.
Table 4.1 Performance of the NIR-red algorithms for the estimation of chl-a
concentrations up to 100 mg m-3.
Algorithm
MERIS three-band algorithm
MERIS two-band algorithm
MODIS two-band algorithm
Enhanced three-band index
Gons’ algorithm
Gons’ algorithm with re-parameterized p and a*φ(λ)
Advanced MERIS three-band algorithm
Advanced MERIS two-band algorithm
(Eq. 4.4)
(Eq. 4.5)
(Eq. 4.6)
(Eq. 4.7)
(Eq. 4.8)
(Eq. 4.9)
(Eq. 4.12)
(Eq. 4.13)
MAE
mg m-3
RMSE
mg m-3
MNAE
%
2.5
2.3
4.6
3.0
5.9
2.3
3.4
2.8
3.3
3.6
6.1
4.0
8.6
3.6
5.8
5.0
18.0
11.6
27.6
18.5
26.7
12.5
15.6
12.6
MAE – mean absolute error, RMSE – root mean squared error, and MNAE – mean normalized absolute
error
Table 4.2 Performance of the NIR-red algorithms for the estimation of chl-a
concentrations up to 25 mg m-3.
Algorithm
MERIS three-band algorithm
MERIS two-band algorithm
MODIS two-band algorithm
Enhanced three-band index
Gons’ algorithm
Gons’ algorithm with re-parameterized p and a*φ(λ)
Advanced MERIS three-band algorithm
Advanced MERIS two-band algorithm
(Eq. 4.4)
(Eq. 4.5)
(Eq. 4.6)
(Eq. 4.7)
(Eq. 4.8)
(Eq. 4.9)
(Eq. 4.12)
(Eq. 4.13)
MAE
mg m-3
RMSE
mg m-3
MNAE
%
1.9
1.2
3.1
2.5
3.1
1.3
1.4
1.2
2.1
1.4
3.8
2.9
3.5
1.5
1.6
1.6
21.5
11.5
30.9
22.4
26.6
13.3
15.2
11.9
93
4.6
Validity of the developed algorithms in turbid highly
productive inland waters
The relationships of chl-a concentrations and model values identified through the
calibration of the MERIS three-band and two-band models increased in linearity with an
increase in chl-a concentration and changed to linear relationships for chl-a
concentrations above 15 to 20 mg m-3 (Figs. 4.22 and 4.23). This may result in an
overestimation of chl-a concentrations if the algorithms developed for the estimation of
chl-a concentrations from 2.3 to 81.2 mg m-3 are applied for waters with substantially
higher chl-a concentrations. The relationship of chl-a concentrations and model values
identified for the MODIS two-band model remained linear with an increase in chl-a
concentration and this algorithm is expected to estimate high chl-a concentrations with a
reasonable accuracy (Fig. 4.24).
3-band Model - MERIS
0.75
Fremont Lakes 2008
Branched Oak Lake 2011
0.50
0.25
0.00
y = 0.00381x - 0.10589
R² = 0.95
-0.25
0
25
50
75
100
Chl-a, mg
125
150
175
m-3
Figure 4.22 Relationship of chl-a concentrations and MERIS three-band model values for
the Fremont Lakes 2008 dataset, used for the algorithm development, and the Branched Oak
Lake 2011 dataset. The linear regression line is representative of the combined dataset.
(ρ665.00)-1 x ρ708.75
94
3.5
Fremont Lakes 2008
3.0
Branched Oak Lake 2011
2.5
2.0
1.5
1.0
y = 0.01376x + 0.64225
R² = 0.97
0.5
0.0
0
25
50
75
100
Chl-a, mg m-3
125
150
175
Figure 4.23 Relationship of chl-a concentrations and MERIS two-band model values for
the Fremont Lakes 2008 and Branched Oak Lake 2011 datasets. The linear regression
line is representative of the combined dataset.
1.25
Fremont Lakes 2008
Branched Oak Lake 2011
(ρ667)-1 x ρ748
1.00
Fremont Lakes 2008
y = 0.00394x + 0.20867
R² = 0.75
0.75
0.50
Branched Oak Lake 2011
y = 0.00387x + 0.20664
R² = 0.91
0.25
0.00
0
25
50
75
100
Chl-a, mg
125
150
175
m-3
Figure 4.24 Relationship of chl-a concentrations and MODIS two-band model values for
the Fremont Lakes 2008 and Branched Oak Lake 2011 datasets. The offsets and
intercepts of the regression lines for the Fremont Lakes 2008 dataset and the combined
dataset are practically identical.
95
4.7
Estimation of phycocyanin with Simis’ red-NIR algorithm
We applied the semi-empirical algorithm, developed by Simis et al. (2005) for
MERIS band settings (Eqs. 4.14 to 4.16), for the estimation of phycocyanin
concentrations for a small number of samples taken and analyzed for the Fremont Lakes
in 2009 (Tab. 4.3). The estimation of phycocyanin with this model involved three
different steps.
Firstly, the absorption coefficient of chl-a at 665 nm, achl-a(665), was estimated to
correct for possible absorption by chl-a in the spectral region where phycocyanin absorbs:
  ρ (709) 
 −1


achl−a (665) =  
×
[
(
a
(
7
09
)
+
b
)
]
a
(
6
65
)
b
w
b 
w
b ×γ
  ρ (665) 



(4.14)
where ρ(665) and ρ(709) are the simulated reflectances for MERIS bands 7 (centered at
665.00 nm) and 9 (centered at 708.75 nm), aw(665) and aw(709), the absorption
coefficients of water for MERIS bands 7 and 9, were set to 0.4 and 0.7 (from Buiteveld,
1994), bb, the backscattering coefficient, was derived from MERIS band 12 (centered at
778.75 nm), and γ, a correction factor to correct for differences in the estimated and
analytically measured aφ(665), was set to 0.68.
Secondly, the absorption coefficient of phycocyanin at 620 nm, aPC(620), was
retrieved:
  ρ (709) 


a PC (620) =  
× [(a w (709) + bb )] - a w (620) - bb  × δ −1 − [ε × achl-a (665)]

  ρ (620) 




(4.15)
where ρ(620) and ρ(709) are the simulated reflectances for MERIS bands 6 (centered at
620.00 nm) and 9 (centered at 708.75 nm), aw(620) and aw(709), the absorption
96
coefficients of water for MERIS bands 6 and 9, were set to 0.3 and 0.7 (from Buiteveld,
1994), bb was derived from MERIS band 12 (centered at 778.75 nm), and δ, a correction
factor to correct for differences in the estimated and analytically measured aφ(620), and
was set to 0.84. The conversion factor ε, introduced to relate the absorption by chl-a at
665 nm to its absorption at 620 nm, was set to 0.24.
The concentration of phycocyanin was retrieved through the division of aPC(620)
by the phycocyanin specific absorption coefficient of phytoplankton, a*PC(620):
[PC] =
a PC (620)
∗
a PC
(620)
(4.16)
where a*PC(620) was set to 0.0095 m2 mg-1.
Table 4.3 Descriptive statistics of the water quality parameters
N
Min
Max
Median
Mean
Standard
Deviation
CV
(%)
(mg m-3)
15 10.5
196.4
33.0
53.1
50.1
94.3
Chl-a
(mg m-3)
15
0.7
159.4
7.6
25.1
41.1
164.1
PC
15
0.0035
1.5569
0.3795
0.4090
0.3821
93.4
PC/Chl-a
(m)
15
0.44
1.86
0.84
0.95
0.46
48.5
SD depth
(NTU)
15
2.32
23.40
8.15
9.35
6.03
64.5
Turbidity
15
2.40
18.11
7.68
9.16
5.32
58.0
(g m-3)
TSS
(g m-3)
15
0.17
6.22
1.00
1.45
1.57
108.4
ISS
(g m-3)
15
1.80
17.00
6.83
7.71
4.56
59.1
OSS
(m-1)
15
0.0407
1.1386
0.2740
0.3984
0.3515
88.2
aφ(620)
(m-1)
15
0.0727
1.9790
0.3493
0.5931
0.5586
94.2
aφ(665)
*
(m2 mg-1) 15
0.0070
0.9469
0.0193
0.0938
0.2376
253.4
a PC(620)
(m2 mg-1) 15
0.0069
0.0139
0.0110
0.0108
0.0019
17.3
a*chl-a(665)
Chl-a – Fluorometrically measured chlorophyll-a concentration, PC –phycocyanin concentration, SD depth
– Secchi disk depth, TSS - concentration of total suspended solids, ISS - concentration of inorganic
suspended solids, OSS - concentration of organic suspended solids, aφ(620) and aφ(665) - absorption
coefficients of phytoplankton at 620 and 665 nm, a*PC(620) and a*chl-a(665) - phycocyanin specific and
chl-a specific absorption coefficients of phytoplankton at 620 and 665 nm, N - number of samples, and CV
- coefficient of variation.
97
The algorithm for the estimation of achl-a(665), (Eq. 4.14), contains the correction
factor γ, which represents the ratio of the estimated to the measured absorption coefficient
of phytoplankton, achl-a(665), and reduces the effects of possible differences in the
absorption by NAP and CDOM at λ1 and λ2. We compared the ratio of the measured
achl-a(665) to the estimated achl-a(665) for the Fremont Lakes 2008 dataset (Fig. 4.25) and
found a substantially different correction factor γ of 0.9986 from Simis et al. (2005). The
γ of 0.68 reported by Simis et al. (2005) was, therefore, replaced by the γ of 0.9986.
The validity of the γ retrieved from the Fremont Lakes 2008 dataset for the
estimation of aφ(665) for the Fremont Lakes 2009 dataset was confirmed, when the
estimated aφ(665) was divided by the median analytically measured a*chl-a(665) value of
0.0112 for the Fremont Lakes 2008 dataset, and chl-a was estimated very accurately (Fig.
4.26). The algorithm for the retrieval of aPC(620), was applied with its original
parameterization (Eq. 4.15) and a*PC(620) was taken from Simis et al. (2005).
Estimated aφ(665), nm-1
1.25
γ = 0.9986
1.00
0.75
0.50
0.25
0.00
0.00
y = 0.9986x
R² = 0.8063
0.25
0.50
0.75
1.00
Measured aφ(665), mg m-3
1.25
Figure 4.25 Identification of γ in Eq. 4.14 through a linear least squares fit of the
measured and estimated (Eq. 4.14 with γ = 1) aφ(665). The intercept was set to zero.
98
Chl-aestimated, mg m-3
100
75
50
MAE = 2.3 mg m-3
RMSE = 3.6 mg m-3
MNAE = 13.1 %
25
0
0
25
50
75
Chl-aobserved, mg m-3
100
Figure 4.26 Application of a variation of Gons’ algorithm (Eq. 4.14) for the estimation of
chl-a concentrations up to 100 mg m-3 for the Fremont Lakes 2009 dataset. The algorithm
was re-parameterized with γ = 0.9986 and the median analytically measured a*chl-a(665)
value of 0.0112 for the Fremont Lakes 2008 dataset.
The estimation of the phycocyanin concentrations by the algorithm parameterized
for Lake Loosdrecht and Lake Ijsselmeer in the Netherlands (Simis et al., 2005) resulted
in a substantial overestimation of the phycocyanin concentrations for the Fremont Lakes
2009 dataset (Fig. 4.27) comparable to the results for Lake Ijsselmeer when a fixed value
of a*PC(620) was applied (Simis et al., 2005). Despite overestimation, the estimated
phycocyanin concentrations showed a close relationship with the measured phycocyanin
concentrations, with the exception of two stations with very different bio-physical and
bio-optical conditions. These two outliers were expected.
The first station had a fluorometrically measured chl-a concentration of 196.4 mg
m-3 (the spectrophotometrically measured chl-a concentration was practically identical with
195.2 mg m-3) and a phycocyanin concentration of only 0.69 mg m-3. The phytoplankton
composition for this station was dominated by Chlorophyta and Bacillariophyta,
99
indicated by the relatively high chl-b and –c concentrations of 12.6 and 7.0 mg m-3, and
interferences from absorption by chl-b and chl-c with the estimation of aPC(620) were
expected.
The second station had a fluorometrically measured chl-a concentration of 102.4
mg m-3 (the spectrophotometrically measured chl-a concentration was 102.8 mg m-3) and
a phycocyanin concentration of 159.4 mg m-3 and was the only station with a PC/Chl-a
ratio above 1. The data collection at this station coincided with a Cyanobacteria bloom
and the phytoplankton composition was dominated by Cyanobacteria. The concentration
of chl-b, a pigment typically present in the early summer, was smaller than 0.1 mg m-3
and the chl-c concentration was only 2.8 mg m-3.
Phycocyaninestimated, mg m-3
200
1:1 Line
y = 1.57x + 7.92
R² = 0.89
150
100
50
MAE = 17.4 mg m-3
RMSE = 22.4 mg m-3
MNAE = 267.9 %
0
0
50
100
150
Phycocyaninobserved, mg m-3
200
Figure 4.27 Application of Simis’ algorithm (Eqs. 4.15 and 4.16) for the estimation of
phycocyanin concentrations for the Fremont Lakes 2009 dataset. The two outliers marked
in red were excluded from the calculation of the statistics.
100
We compared the measured aφ(620) and the estimated aφ(620) for the Fremont
Lakes 2009 dataset to investigate the overestimation of the phycocyanin concentrations
(Fig. 4.28) and found a very close relationship with a slightly different correction factor δ
(Eq. 4.15) compared to Simis et al. (2005). The reason for the overestimation of
phycocyanin is, therefore, likely related to the high variability in a*PC(620) combined
with the errors in the estimation of the contribution of achl-a(620) to aφ(620) and the
potential contribution of absorption by pigments different from chl-a to aφ(620) if the
phytoplankton is only partially composed by Cyanobacteria (Simis et al., 2005, Simis et
al., 2007). These conditions are typical for the Fremont Lakes and the seasonal
differences in the contribution of Cyanobacteria to the phytoplankton composition in the
Fremont Lakes are reflected in the inverse relationship of PC/Chl-a ratio and a*PC(620)
shown in Fig. 4.29.
Estimated aφ(620), nm-1
1.25
δ = 0.9526
1.00
0.75
0.50
0.25
y = 0.9526x
R² = 0.8920
0.00
0.00
0.25
0.50
0.75
1.00
1.25
Measured aφ(620), mg m-3
Figure 4.28 Identification of δ in Eq. 4.15 through a linear least squares fit of the
measured and estimated (Eq. 4.15 with δ = 1) aφ(620). The intercept was set to zero.
101
aPC*(620), nm-1
0.10
R² = 0.87
0.05
0.00
0.00
0.25
0.50
0.75
PC/Chl-a
Figure 4.29 Relationship of PC/Chl-a and analytically measured phycocyanin specific
absorption coefficient of phytoplankton, a*PC(620). The relationship is different from the
relationship expected for Cyanobacteria dominated waters for which a relatively constant
a*PC(620) is typical. The two previously identified outliers have PC/Chl-a ratios of
0.0035 and 1.5568 and a a*PC(620) of 0.0070 and 0.9469 and fit the relationship shown.
The algorithm for the estimation of aPC(620), (Eq. 4.15), contains the conversion
factor ε to relate the absorption by chl-a at 620 nm to its absorption at 665 nm. We
retrieved ε trough iterative linear regressions of the measured vs. the estimated
phycocyanin concentrations and the optimal value of ε was reached when the slope of the
relationship reached 1 (Simis et al., 2005). The retrieved value of ε of 0.57 resulted in a
a*PC(620) of 0.0078. This a*PC(620) value seems typical for inland waters dominated by
Cyanobacteria (Simis et al., 2007) and was only slightly different from the analytically
measured a*PC(620) of 0.0070 for the only station with a PC/Chl-a ratio above 1. The
conversion factor ε, found in this study, was very different from the value of 0.24 found
by Simis et al. (2005) for the Cyanobacteria dominated Lake Loosdrecht and reflected the
102
higher contribution of the absorption by chl-a to aφ(620) in the Fremont Lakes.
Therefore, the algorithm for the estimation of aPC(620), (Eq. 4.15), was re-parameterized
with the found δ of 0.9526 and the revised conversion factor ε of 0.57.
The estimation of the phycocyanin concentrations after the re-parameterization of
δ, ε, and a*PC(620) still resulted in a substantial overestimation of the phycocyanin
concentrations for some stations with PC/Chl-a ratios of up to 0.2 while the algorithm
worked more accurately for the rest of the Fremont Lakes 2009 dataset (Fig. 4.30). The
effects of a high variability in a*PC(620), insufficient removal of the contribution of achla(620)
to aPC(620) and potential interferences with the absorption by pigments different
from chl-a resulted in estimation errors of up to 327.0% for these stations and of up to
64.6% for the rest of the stations and require continued investigation.
Phycocyaninmodeled, mg m-3
75
1:1 Line
y = 1.00x + 0.03
R² = 0.84
50
25
MAE = 5.6mg m-3
RMSE = 7.7 mg m-3
MNAE = 64.4 %
0
0
25
50
Phycocyaninobserved, mg
75
m-3
Figure 4.30 Application of Simis’ re-parameterized algorithm (Simis et al., 2005) for the
estimation of phycocyanin concentrations for the Fremont Lakes 2009 dataset. The two
outliers were excluded.
103
Chapter 5. Remote Estimation of Chl-a in Turbid
Productive Waters from Multi-temporal AISA Imagery
5.1
Bio-physical and bio-optical water quality parameters of the
Fremont Lakes for the AISA data collection dates
The applicability of NIR-red models for the remote estimation of chl-a
concentrations was studied for five hyperspectral AISA images taken on 07/02/2008,
07/14/2008, 09/26/2008, 10/25/2008, and 11/19/2008 at the Fremont Lakes. This dataset
represented a part of the Fremont Lakes 2008 dataset and showed slightly different biophysical and bio-optical conditions compared to the Fremont Lakes 2008 dataset.
The descriptive statistics of the water quality parameters for this dataset (Tab. 5.1)
confirm the high variability in the bio-physical and bio-optical conditions typical for the
Fremont Lakes. The chl-a concentrations ranged from 2.3 to 200.8 mg m-3 and the TSS
concentrations ranged from 1.19 to 15.00 g m-3. The median chl-a concentration of 23.3
mg m-3 measured for this dataset (Tab. 5.1) was slightly different from the median chl-a
concentration of 27.4 mg m-3 measured for the Fremont Lakes 2008 dataset (Tab. 3.1).
This difference was not reflected in the TSS concentrations and the median TSS
concentration was 6.86 g m-3 for this dataset (Tab. 5.1) compared to 6.80 g m-3 for the
Fremont Lakes 2008 dataset (Tab. 3.1). ISS concentrations ranged from 0.2 to 3.5 g m-3
and the contribution of ISS to TSS ranged from 4.3 to 58.7% with a median contribution
of 16.0%. OSS concentrations ranged from 0.8 to 12.5 g m-3.
104
Table 5.1 Descriptive statistics of the water quality parameters measured at the times of
AISA data collection in 2008.
N
Min
Max
Median
Mean
Standard
Deviation
CV
(%)
(mg m-3)
37
2.27
200.81
23.32
34.57
38.93
112.6
Chl-a
(m)
37
0.51
4.20
1.08
1.37
0.88
63.9
SD depth
(NTU)
37
1.51
19.20
6.73
7.75
5.11
65.9
Turbidity
(g m-3)
37
1.19
15.00
6.86
7.38
3.75
50.8
TSS
(g m-3)
37
0.22
5.85
1.09
1.42
1.27
88.9
ISS
(g m-3)
37
0.81
12.50
5.85
5.95
3.22
54.1
OSS
(m-1)
37
0.0470
4.4867
0.4643
0.6557
0.7859
119.8
ap (675)
(m-1)
37
0.0031
0.2317
0.0503
0.0572
0.0458
80.1
aNAP (675)
37
0.0440
4.4051
0.4158
0.5986
0.7776
129.9
(m-1)
aφ (675)
(m2 mg-1)
37
0.0074
0.0285
0.0168
0.0181
0.0057
31.3
a*φ(675)
37
0.4668
1.4525
0.8309
0.8796
0.2611
29.7
aCDOM (440) (m-1)
37
0.0069
0.0333
0.0152
0.0164
0.0061
37.3
aCDOM (675) (m-1)
(nm-1)
37
0.0156
0.0185
0.0172
0.0170
0.0007
4.0
SCDOM
Chl-a – Fluorometrically measured chlorophyll-a concentration, SD depth – Secchi disk depth, TSS concentration of total suspended solids, ISS - concentration of inorganic suspended solids, OSS concentration of organic suspended solids, ap(675) - total particulate absorption coefficient at 675 nm,
aNAP(675) - absorption coefficient of non-algal particles at 675 nm, aφ(675) - absorption coefficient of
phytoplankton at 675 nm, a*φ(675) - chl-a specific absorption coefficient of phytoplankton at 675 nm,
aCDOM(440) and aCDOM(675) - absorption coefficients of CDOM at 440 and 675 nm, N - number of
samples, and CV - coefficient of variation.
The spectra of ap(λ), aNAP(λ), aφ(λ), and aCDOM(λ) showed a high variability and
were representative for the bio-physical and bio-optical conditions found in the Fremont
Lakes in 2008. The median SCDOM was practically identical for the two datasets. The
seasonal differences in the chl-a concentrations, and possibly in the species composition
of the phytoplankton in the Fremont Lakes, resulted in a different a*φ(675) for this dataset
(Tab. 5.1) and the Fremont Lakes 2008 dataset (Tab. 3.1). The median a*φ(675) for this
dataset was 0.0168 m2 mg-1 compared to 0.0143 m2 mg-1 for the Fremont Lakes 2008
dataset and represented a station with a chl-a concentration of 25.25 mg m-3 (Fig. 5.1).
105
Fremont Lakes 2008 datset
0.050
a*φ (675), m2 mg-1
Fremont Lakes 2008 dataset for AISA
0.025
0.000
0
50
100
Chl-a, mg m-3
150
200
Figure 5.1 Chl-a specific absorption coefficient of phytoplankton at 675 nm, a*φ(675),
shown versus chl-a concentration.
5.2
Comparison of AISA reflectance spectra and in situ
reflectance spectra
The QUick atmospheric correction (Bernstein et al., 2005a,b), QUAC, of the
AISA images preserved the typical reflectance features of turbid productive waters even
though there was an offset in the magnitude of the AISA reflectance spectra when
compared to the magnitude of the in situ reflectance spectra measured with the Ocean
Optics® USB2000 spectrometers (Moses et al., 2012). Possible reasons for this offset
include the incomplete removal of atmospheric effects in the atmospheric correction
process, specular reflectance from the water surface, and differences in the radiometric
characteristics of the AISA sensor and the Ocean Optics® USB2000 spectrometers.
The reflectance in the blue spectral region was relatively low and without any
distinctive spectral features (Fig. 5.2). The reflectance increased in the green spectral
region and typically reached a maximum at 563 to 572 nm in consequence of the minimal
106
absorption by phytoplankton pigments. Several reflectance spectra showed a trough,
related to the absorption by phycocyanin, at 620 to 630 nm. The reflectance in the red and
NIR spectral regions was characterized by a trough at 675 nm, which is caused by
absorption by chl-a (Bidigare et al., 1990), and the peak around 700 nm, related to the
minimum in the combined absorption by phytoplankton pigments and water (Vasilkov
and Kopelevich, 1982; Gitelson, 1992 and 1993).
1.0
ρrs (λ), sr-1
0.04
0.03
0.5
0.02
0.01
0.00
Coefficient of Variation
0.05
0.0
400
500
600
700
Wavelength, nm
800
900
Figure 5.2 Remote sensing reflectance, ρrs(λ), spectra derived from the AISA
measurements. The coefficient of variation of the reflectance is shown in grey.
The magnitude of the reflectance was highly variable with coefficients of
variation from 23.3 to 70.6%. Some of this variability, especially in the blue region,
seemed caused by the incomplete removal of path radiance in the atmospheric correction
process, while the variability in the NIR region, at wavelengths from 730 to 800 nm, was
mainly caused by the effects of absorption by atmospheric water vapor. The chl-a
specific absorption features in the red and NIR regions seemed well preserved for stations
with a variety of chl-a concentrations (Fig. 5.3).
107
0.006
0.009
0.004
0.006
0.002
0.003
0.000
In situ ρrs(λ), sr-1
QUAC ρrs (λ), sr-1
a 0.012
0.000
400
500
600
700
800
900
Wavelength, nm
0.003
0.004
0.002
0.002
0.001
0.000
0.000
400
500
600
700
Wavelength, nm
800
900
c 0.018
QUAC ρrs (λ), sr-1
In situ ρrs(λ), sr-1
0.004
0.012
0.009
0.012
0.006
0.006
0.003
0.000
In situ ρrs(λ), sr-1
QUAC ρrs (λ), sr-1
b 0.006
0.000
400
500
600
700
Wavelength, nm
800
900
Figure 5.3 Comparison of the QUAC corrected AISA spectra (solid lines) and the in situ
measured Ocean Optics spectra (dashed lines) for three stations with chl-a concentrations
of 7.6 mg m-3 (a), 25.2 mg m-3 (b), and 81.2 mg m-3 (c).
108
5.3
Calibration of the NIR-red models for the estimation of chl-a
concentrations from AISA imagery
Relationships of chl-a concentrations and model values were established for the
QUAC corrected AISA images taken on 07/14/2008 and 10/25/2008. These datasets
comprised 12 stations with chl-a concentrations from 7.6 to 80.2 mg m-3. Stations with
chl-a concentations above 100 mg m-3 were excluded from the calibration and validation
of the models.
The calibration of the NIR-red models for only two datasets made it possible to
study the applicability of the developed algorithms for the three datasets taken on
07/02/2008, 09/26/2008, and 11/19/2008. These datasets are potentially affected by
differences in the atmospheric correction results for the individual images and seasonal
differences in the species composition of the phytoplankton and AISA imagery might
provide a powerful tool for the evaluation of the trophic state of inland waters if it was
possible to apply the algorithms developed for a relatively small dataset to an
independent multi-temporal dataset.
The three-band model (Eq. 1.5) and two-band model (Eq. 1.7), which were
successfully calibrated and validated within the spectral bands of MERIS for the Fremont
Lakes 2008 and 2009 datasets, were calibrated within the spectral bands of AISA. The
band positions for λ1 and λ2 for the three-band model (Eq. 1.5) and two-band model (Eq.
1.7) were set to match MERIS bands 7 and 9 whereas λ3 for the three-band model (Eq.
1.5) was set to a shorter wavelength compared to MERIS band 10 due to the effects of
instrument noise in the NIR spectral region, at wavelengths of 730 nm and beyond, where
109
some of the atmospheric absorption bands of water vapor are found. The final
formulations of the AISA three-band model and two-band model were:
AISA three-band model = [ρrs-1(666) – ρrs-1(704)] × ρrs(723)
(5.1)
AISA two-band model = ρrs-1(666) × ρrs(704)
(5.2)
where ρrs-1(666) is the inverse reflectance for AISA band 30 (centered at 666.39 nm),
ρrs-1(704) is the inverse reflectance for AISA band 34 (centered at 702.02 nm), and
ρrs(723) is the reflectance for AISA band 36 (centered at 722.88 nm) for the sensor
configuration flown at the Fremont Lakes in 2008.
For chl-a concentrations from 7.6 to 80.2 mg m-3, the AISA three-band model
(Eq. 5.1) and AISA two-band model (Eq. 5.2) showed very close relationships (R2 =
0.94) with chl-a concentration, which were practically identical with linear relationships
(Fig. 5.4a and 5.4b). The algorithms for the remote estimation of chl-a concentrations
from the QUAC corrected AISA imagery, identified through the calibration of the
models, were:
Chl-a = 34.98 × (AISA 3BM)2 + 113.69 × (AISA 3BM) + 24.12
(5.3)
for the AISA three-band model (Eq. 5.1) and
Chl-a = 16.84 × (AISA 2BM)2 + 48.11 × (AISA 2BM) – 41.03
for the AISA two-band model (Eq. 5.2).
(5.4)
110
These algorithms were very different from the algorithms developed for MERIS
(Eq. 4.4 and 4.5) and seemed to represent stations with low-to-moderate chl-a
concentrations even more effectively. The median chl-a concentration for this dataset was
only 23.3 mg m-3 compared to 26.9 mg m-3 for the dataset used to calibrate the NIR-red
models for MERIS and MODIS for chl-a concentrations from 2.3 to 81.2 mg m-3.
3-band Model - AISA
a 0.50
0.25
0.00
0
20
40
60
80
100
-0.25
-0.50
R² = 0.94
Chl-a, mg m-3
2-band Model - AISA
b 2.00
1.50
1.00
0.50
R² = 0.94
0.00
0
25
50
75
100
Chl-a, mg m-3
Figure 5.4 Calibration of the Eq. 5.1 AISA three-band model (a), and Eq. 5.2 AISA twoband model (b) for the estimation of chl-a concentrations from the QUAC corrected
AISA images.
111
5.4
Validation of the NIR-red models for the estimation of chl-a
concentrations for AISA imagery
The developed algorithms were validated for the QUAC corrected AISA images
taken on 07/02/2008, 09/26/2008, and 11/19/2008. Therefore, the chl-a concentrations
were estimated by Eqs. 5.3 and 5.4 and the estimated chl-a concentrations, chl-aestimated,
were compared with the observed chl-a concentrations, chl-aobserved.
The results of the validation (Figs. 5.5 and 5.6, Tab. 5.2) were only slightly
different for the AISA three-band and two-band algorithms (Eqs. 5.3 and 5.4). The mean
absolute error, MAE, for the estimation of chl-a concentrations from 2.3 to 81.2 mg m-3
was 5.9 mg m-3 for the AISA three-band algorithm and 5.5 mg m-3 for the AISA twoband algorithm and the relatively small difference in the MAE was statistically
insignificant (Student’s t-test, p = 0.54).
The AISA three-band algorithm (Eq. 5.3) overestimated the chl-a concentrations
for some stations for 07/02/2008 and slightly underestimated the chl-a concentrations for
09/26/2008 and 11/19/2008. There was a strong linear relationship of observed and
estimated chl-a concentrations for the three dates (R2 = 0.95) and some of the spread of
the points from the 1:1 line is probably a result of the differences in the residual
atmospheric effects for the individual images. These atmospheric effects were probably
relatively small at λ1 and λ2, for which the chl-a specific absorption features in the red
and NIR regions seemed well preserved, and increased at λ3, which was much more
affected by the instrument noise in the NIR spectral region (Fig. 5.3). This may explain
112
the slightly more accurately estimated chl-a concentrations for the AISA two-band
algorithm (Eq. 5.4).
Chl-aestimated, mg m-3
100
3-Band Model - AISA
1:1 Line
07/02/2008
75
09/26/2008
11/19/2008
50
MAE = 5.2 mg m-3
RMSE = 5.9 mg m-3
MNAE = 39.9 %
25
0
0
25
50
Chl-aobserved, mg
75
100
m-3
Figure 5.5 Application of the AISA three-band model, calibrated from the QUAC
corrected AISA images taken on 07/14/2008 and 10/25/2008, for the estimation of chl-a
concentrations for the images taken on 07/02/2008, 09/26/2008, and 11/19/2008.
Chl-aestimated, mg m-3
100
1:1 Line
2-Band Model - AISA
07/02/2008
75
09/26/2008
11/19/2008
50
MAE = 4.4 mg m-3
RMSE = 5.5 mg m-3
MNAE = 30.3 %
25
0
0
25
50
Chl-aobserved, mg
75
100
m-3
Figure 5.6 Application of the AISA two-band model, calibrated from the QUAC
corrected AISA images taken on 07/14/2008 and 10/25/2008, for the estimation of chl-a
concentrations for the images taken on 07/02/2008, 09/26/2008, and 11/19/2008.
113
5.5
Comparison of Gon’s model and the MERIS two-band model
for the estimation of chl-a concentrations for AISA imagery
The comparison of the performance of Gons’ model (Eq. 1.2) to the performance
of the AISA two-band model (Eq. 5.2) required the development of an algorithm
specifically designed for the AISA sensor, and we, therefore, calibrated the model for the
QUAC corrected AISA images taken on 07/14/2008 and 10/25/2008. The formulation of
the model for the AISA sensor was:
[Chl - a] =
ρ (704)
p
(a w (704) + bb ) - a w (666) - bb
ρ (666)
*
a φ (666)
(Eq. 5.5)
where ρrs(666) is the reflectance for AISA band 30 (centered at 666.39 nm) and ρrs(704)
is the reflectance for AISA band 34 (centered at 702.02 nm) for the sensor configuration
flown at the Fremont Lakes in 2008, aw(666) and aw(704) were set to 0.4 and 0.7, and bb
was retrieved from AISA band 42 (centered at 780.01 nm).
The values for aw(666.39) and aw(704.02) were interpolated from Mueller (2003)
and the coefficients for a*φ(666.39) and p were found through minimizing the STE for the
calibration dataset. The identified a*φ(666.39) of 0.0138 m-1 differed substantially from
the median a*φ(666) of 0.0114 m-1 we measured in the laboratory for the calibration
dataset and was more similar to the average a*φ(666) of 0.0136 m-1 instead. This resulted
in an increase in the sensitivity of the algorithm toward stations with a relatively high
a*φ(666). The exponent p was set to 1.052.
114
We found a linear relationship (Fig. 5.7) between the estimated and observed
chl-a concentrations when we validated the developed algorithm for the QUAC corrected
AISA images taken on 07/02/2008, 09/26/2008, and 11/19/2008. The MAE for the
estimation of chl-a concentrations from 2.3 to 81.2 mg m-3 of 4.6 mg m-3 for this
algorithm was comparable to the MAE of 4.4 mg m-3 for the two-band algorithm (Tab.
5.2). Similarly to the AISA three-band algorithm (Eq. 5.3) and AISA two-band algorithm
(Eq. 5.4), the algorithm overestimated the chl-a concentrations for some stations for
07/02/2008 and slightly underestimated the chl-a concentrations for 09/26/2008 and
11/19/2008.
Chl-aestimated, mg m-3
100
1:1 Line
Gons' Model - AISA
07/02/2008
75
09/26/2008
11/19/2008
50
MAE = 4.6 mg m-3
RMSE = 5.6 mg m-3
MNAE = 28.9 %
25
0
0
25
50
Chl-aobserved, mg
75
100
m-3
Figure 5.7 Application of the Gons’ model, calibrated from the QUAC corrected AISA
images taken on 07/14/2008 and 10/25/2008, for the estimation of chl-a concentrations
for the images taken on 07/02/2008, 09/26/2008, and 11/19/2008.
115
These results are probably related to differences in a*φ(666). Even a relatively
small difference in a*φ(666) for the calibration and validation datasets causes
considerable changes in the slope of the relationship of observed and estimated chl-a
concentrations and the a*φ(666) of 0.0138 used for the calibration of the model was
substantially lower compared to the median a*φ(666) of 0.0181 we measured in the
laboratory for 07/02/2008, which caused the overestimation of the chl-a concentrations
for this date, and slightly higher compared to median a*φ(666) of 0.0127 for 09/26/2008
and 11/19/2008, which resulted in the underestimation of the chl-a concentrations for
these dates.
Table 5.2 Performance of the NIR-red algorithms for the estimation of chl-a
concentrations from the QUAC corrected AISA imagery
Algorithm
MERIS three-band algorithm
MERIS two-band algorithm
Gons’ algorithm with re-parameterized p and a*φ(λ)
(Eq. 5.3)
(Eq. 5.4)
(Eq. 5.5)
MAE
mg m-3
RMSE
mg m-3
MNAE
%
5.2
4.4
4.6
5.9
5.5
5.6
39.9
30.3
28.9
MAE – mean absolute error, RMSE – root mean squared error, and MNAE – mean normalized absolute
error
Each of the applied algorithms provided reliable estimates of the chl-a
concentrations in the Fremont Lakes for the QUAC corrected AISA images for this
multi-temporal dataset even though the algorithms seemed sensitive to variations in
a*φ(666). The algorithms successfully captured the spatial and temporal variability of the
chl-a concentrations found in the Fremont Lakes in summer and fall 2008. This is
reflected in the maps of the chl-a concentrations estimated by the AISA two-band
algorithm (Eq. 5.4) for 07/02/2008, 09/26/2008, and 11/119/2008 (Figs. 5.8 to 5.10).
116
02 Jul 2008
/
-3
Chl-a
Legend (mg m )
< 10
10 - 20
20 - 30
30 - 40
40 - 50
> 50
0
0.5
1
1.5
Kilometers
Figure 5.8 Map of the chl-a distribution in the Fremont Lakes for 07/02/2008 produced
from the AISA two-band algorithm (Eq. 5.4).
26 Sep 2008
/
-3
Chl-a
Legend (mg m )
< 10
10 - 20
20 - 30
30 - 40
40 - 50
> 50
0
0.5
1
1.5
Kilometers
Figure 5.9 Map of the chl-a distribution in the Fremont Lakes for 09/26/2008 produced
from the AISA two-band algorithm (Eq. 5.4).
117
19 Nov 2008
/
-3
Chl-a
Legend (mg m )
< 10
10 - 20
20 - 30
30 - 40
40 - 50
> 50
0
0.5
1
1.5
Kilometers
Figure 5.10 Map of the chl-a distribution in the Fremont Lakes for 11/19/2008 produced
from the AISA two-band algorithm (Eq. 5.4).
118
Chapter 6. Conclusions and Future Work
This dissertation presented a detailed description of the bio-physical and biooptical water quality parameters measured in the Fremont Lakes and in Branched Oak
Lake in eastern Nebraska, USA. The bio-physical and bio-optical water quality
parameters found in these lakes were comparable to water quality parameters typical for
inland and coastal waters around the world. The effects of the presence of NAP and
CDOM, which make the algorithms for the remote estimation of chl-a concentrations
typically applied for oceanic waters ineffective for optically complex inland and coastal
waters, were presented and the importance of the red and NIR spectral regions for the
development of algorithms for the remote estimation of chl-a concentrations in inland
and coastal waters was shown.
Three NIR-red models were calibrated for a dataset with chl-a concentrations
from 2.3 to 81.2 mg m-3 and validated for independent and statistically significantly
different datasets with chl-a concentrations from 4.0 to 95.5 mg m-3 and 4.0 to 24.2 mg
m-3 for the spectral bands of MERIS and MODIS. The developed MERIS three-band
algorithm estimated chl-a concentrations from 4.0 to 24.2 mg m-3, which are typical for
many inland and coastal waters, very accurately with an MAE of 1.9 mg m-3 and the
developed MERIS two-band algorithm estimated chl-a concentrations even more
accurately with an MAE of 1.2 mg m-3. The MODIS two-band algorithm was less
accurate compared to these algorithms. It estimated chl-a concentrations from 4.0 to 24.2
mg m-3 with an MAE of 3.1 mg m-3.
119
The performance of the MERIS three-band algorithm, MERIS two-band
algorithm, and MODIS two-band algorithm was compared to the performance of several
more complex recently published NIR-red algorithms. Some of these algorithms, the
enhanced three-band index and the advanced MERIS algorithms, estimated chl-a
concentrations similarly accurately to the MERIS three-band algorithm and MERIS twoband algorithm. The MERIS two-band algorithm, the simplest algorithm, provided the
most accurate estimation of chl-a concentrations up to 25 mg m-3. These results indicate
a high potential of the MERIS two-band algorithm for the reliable estimation of chl-a
concentrations in inland and coastal waters, without any reduction in accuracy compared
to the more complex algorithms, even though more research is required to analyze the
sensitivity of the algorithm to differences in a*φ(665).
The performance of a complex NIR-red algorithm applied for the remote
estimation of phycocyanin concentrations in waters with a diverse species composition of
the phytoplankton was investigated for a part of the dataset collected at the Fremont
Lakes. This algorithm estimated phycocyanin concentrations from 1.8 to 59.9 mg m-3
with an MAE of 17.4 mg m-3 when it was applied with its original parameterization for
cyanobacteria dominated inland waters. The estimation of phycocyanin concentrations
improved with an adjusted parameterization for the Fremont Lakes even though the
results were still influenced by the effects of a high variability in a*PC(620), insufficient
removal of the contribution of achl-a(620) to aPC(620), and potential interferences with the
absorption by pigments different from chl-a. This algorithm estimated phycocyanin
concentrations with an MAE of 5.6 mg m-3 when it was applied with its adjusted
parameterization for the Fremont Lakes.
120
Three NIR-red models were calibrated and validated for the remote estimation of
chl-a concentrations in the Fremont Lakes from atmospherically corrected multi-temporal
hyperspectral aerial imagery. The AISA three-band model and AISA two-band model
accurately estimated chl-a concentrations from the hyperspectral aerial imagery. The
developed algorithms estimated chl-a concentrations from 2.3 to 81.2 mg m-3 with MAEs
of 5.2 mg m-3 for the AISA three-band algorithm and 4.4 mg m-3 for the AISA two-band
algorithm. The results from Gons’ model, which was calibrated with the a*φ(666) derived
from the atmospherically corrected multi-temporal hyperspectral aerial imagery, were
comparable to the results from the AISA two-band model with an MAE of 4.5 mg m-3.
The algorithms successfully captured the spatial and temporal variability of the chl-a
concentrations found in the Fremont Lakes and accurately estimated the chl-a
concentrations in the Fremont Lakes even though they were, similarly to the algorithms
developed for the spectral bands of the MERIS and MODIS satellite sensors, sensitive to
differences in a*φ(666).
The use of the red and NIR spectral regions is crucial for the development of
algorithms for the remote estimation of chl-a concentrations in optically complex inland
and coastal waters and the developed NIR-red algorithms accurately estimated chl-a
concentrations in the waters studied. The development of a universally applicable NIRred algorithm for the remote estimation of chl-a concentrations in inland and coastal
waters will require extensive studies of the sensitivity of the presented algorithms to
differences in a*φ(665) for inland and coastal waters in different geographical locations
and with a high variability in the bio-physical and bio-optical water quality parameters.
121
References
Baban, S.-M.-J. (1996) Trophic classification and ecosystem checking of lakes using
remotely sensed information. Hydrological Sciences, 41, 939-957.
Baban, S.-M.-J. (1993) Detecting water quality parameters in Norfolk Broads, UK using
Landsat imagery. International Journal of Remote Sensing, 14, 1247-1267.
Babin, M., Stramski, D., Ferrari, G.-M., Claustre, H., Bricaud, A., Obolensky, G., and N.
Hoepffner (2003) Variations in the light absorption coefficients of phytoplankton,
nonalgal particles, and dissolved organic matter in coastal waters around Europe.
Journal of Geophysical Research, 108, 3211, 20 pp.
Babin, M., Morel, A., and B. Gentili (1996) Remote sensing of sea surface Sun-induced
chlorophyll fluorescence: consequences of natural variations in the optical
characteristics of phytoplankton and the quantum yield of chlorophyll a fluorescence.
International Journal of Remote Sensing, 17, 2417-2448.
Bernstein, L.-S., Adler-Golden, S.-M., Sundberg, R.-L., and Ratkowski, A. (2006)
Improved reflectance retrieval from hyper- and multispectral imagery without prior
scene or sensor information. Proceedings SPIE, Remote Sensing of Clouds and the
Atmosphere XI, 6362, 63622P
Bernstein, L.-S., Adler-Golden, S.-M., Sundberg, R.-L., Levine, R.-Y., Perkins, T.-C.,
Berk, A., Ratkowski, A.-J., Felde, G., and M.-L. Hoke (2005a) A new method for
atmospheric correction and aerosol optical property retrieval for VIS-SWIR multiand hyperspectral imaging sensors: QUAC (QUick atmospheric correction).
Proceedings of the Geoscience and Remote Sensing Symposium, 2005, IGARSS '05,
2005 IEEE International, 5, 3549-3552
Bernstein, L.-S., Adler-Golden, S.-M., Sundberg, R.-L., Levine, R.-Y., Perkins, T.-C.,
Berk, A., Ratkowski, A.-J., Felde, G., and M.-L. Hoke (2005b) Validation of the
Quick atmospheric correction (QUAC) algorithm for VNIR-SWIR multi- and
hyperspectral imagery. Proceedings SPIE, Algorithms and Technologies for
Multispectral, Hyperspectral, and Ultraspectral Imagery XI, 5806, 668-678
Bidigare, R.-R., Ondrusek, M.-E., Morrow, J.-H., and D.-A. Kiefer (1990) In-vivo
absorption properties of algal pigments. Proceedings SPIE 1302, 290-302.
Binding, C.-E., Greenberg, T.-A., R.-P. Bukata (2011) Time series analysis of algal
blooms in Lake of the Woods using the MERIS maximum chlorophyll index. Journal
of Plankton Research, 33, 1847-1852.
122
Binding, C.-E., Greenberg, T.-A., Jerome, J.-H., Bukata, R.-P., anf G. Letourneau (2011)
An assessment of MERIS algal bloom products during an intense bloom in Lake of
the Woods. Journal of Plankton Research, 33, 793-806.
Binding, C.-E., Jerome, J.-H., Bukata, R.-P., and W.-G. Booty (2008) Spectral absorption
properties of dissolved and particulate matter in Lake Erie. Remote Sensing of
Environment, 112, 1702-1711.
Binding, C.-E., Bowers, D.-G., and E.-G. Mitchelson-Jacob (2005) Estimating suspended
sediment concentrations from ocean colour measurements in moderately turbid
waters; the impact of variable particle scattering properties. Remote Sensing of
Environment, 94, 373-383.
Bricaud, A., Babin, M., Morel, A., and H. Claustre (1995) Variability in the chlorophyllspecific absorption coefficients of natural phytoplankton: Analysis and
parameterization. Journal of Geophysical Research, 100, 13321-13332
Bricaud, A., Morel, A., and Prieur, L. (1981) Absorption by dissolved organic matter of
the sea (yellow substance) in the UV and visible domains. Limnology and
Oceanography, 26, 43-53.
Buiteveld, H., Hakwoort, J.-H.-M., and M. Donze (1994) Optical properties of pure
water. Proceedings SPIE, Ocean Optics XII, 2258, 174-183.
Carder, K.-L, Steward, R.-G., Harvey, G.-R., and P.-B. Ortner (1989) Marine humic and
fulvic acids: Their effects on remote sensing of ocean chlorophyll. Limnology and
Oceanography, 34, 68-81.
Carlson, R.-E. (1977) A trophic state index for lakes. Limnology and Oceanography, 22,
361-369
Carmichael, W.-W. (1997) The cyanotoxins. Advances in Botanical Research, 27, 211256.
Carpenter, S.-R., Caraco, N.-F., Correl, D.-L., Howarth, R.-W., Sharpley, A.-N., and V.H. Smith (1998) Nonpoint pollution of surface waters with phosphorus and nitrogen.
Ecological Applications, 8, 559-568
Carpenter, S.-R., Kitchell, J.-F., Cottingham, K.-L., Schindler, D.-E., Christensen, D.-L.,
Post, D.-M., and N. Voichick (1996) Chlorophyll variability, nutrient input, and
grazing: evidence from whole lake experiments. Ecology, 77, 725-735.
Clark, D.-K. (1981) Phytoplankton algorithms for the Nimbus-7 CZCS. In: J.-R.-F.
Gower, ed. Oceanography from Space, Plenum Press, New York, 227-238.
123
Cleveland, J.-S. and Weidemann, A.-D. (1993) Quantifying absorption by aquatic
particles: A multiple scattering correction for glass-fiber filters. Limnology and
Oceanography, 38, 1321-1327.
Costanza, R., d’Arge, R., de Groot, R., Farber, S., Grasso, M., Hannon, B., Limburg, K.,
Naeem, S., O’Neil, R.-V., Paruelo, J., Raskin, R.-G., Sutton, P., and M. van den Belt
(1997) The value of the world’s ecosystem services and natural capital. Nature, 387,
253-260.
Dall’Olmo, G. (2006) Isolation of optical signatures of phytoplankton pigments in turbid
productive waters: remote assessment of chlorophyll-a. PhD dissertation, University
of Nebraska – Lincoln, Lincoln, NE.
Dall’Olmo, G. and Gitelson A.-A. (2005) Effect of bio-optical parameter variability on
the remote estimation of chlorophyll-a concentration in turbid productive waters:
experimental results. Applied Optics, 44, 412-422.
Dall’Olmo, G., Gitelson, A.-A., & Rundquist, D.-C. (2003) Towards a unified approach
for the remote estimation of chlorophyll-a in both terrestrial vegetation and turbid
productive waters. Geophysical Research Letters, 30, 1938, 4 pp.
Dekker, A.-G. (1993) Detection of optical water quality parameters for eutrophic waters
by high resolution remote sensing. Doctoral thesis, Vrije Universiteit, Amsterdam.
Dekker, A.-G., Malthus, T.-J., and E. Seyhan (1991) Quantitative modeling of inland
water quality for high resolution systems. IEEE Transactions on Geoscience and
Remote Sensing, 29, 89-95.
Eaton, A.-D., Clesceri, L.-S., Rice, E.-W., Greenberg, A.-E., and Franson, M.-A.-H.
(Eds.), (2005) Standard methods for the examination of water and wastewater:
centennial edition. American Public Health Association, Baltimore, MD: United
Book Press, Inc., (Sections 2540 D. Total suspended solids dried at 103 – 105°C and
2540 E. Fixed and volatile solids ignited at 550°C.).
Effler, S.-W., Perkins, M., Peng, F., Strait, C., Weidemann, A.-D., and Auer, T. (2010)
Light-absorbing components in Lake Superior. Journal of Great Lakes Research, 36,
656-665
Ferrari, G.-M. and Tassan, S. (1999) A method using chemical oxidation to remove light
absorption by phytoplankton pigments. Journal of Phycology, 35, 1090-1098.
Gilerson, A., Gitelson, A.-A., Zhou, J., Gurlin, D., Moses, W.-J., Ioannou, I., and Ahmed,
S.-A. (2010) Algorithms for remote estimation of chlorophyll-a in coastal and inland
waters using red and near infrared bands. Optics Express, 18, 24109-24125.
124
Gitelson, A.-A., Gurlin, D., Moses, W., and Yacobi, Y. (2011) Remote Estimation of
Chlorophyll-A Concentration in Inland, Estuarine, and Coastal Waters. In: Q. Weng,
ed. Advances in Environmental Remote Sensing: Sensors, Algorithms, and
Applications, chapter 18, CRC Press, Taylor and Francis Group, 449-478.
Gitelson, A.-A., Dall’Olmo, G., Moses, W., Rundquist, D., Barrow, T., Fisher, T.-R.,
Gurlin, D., and J. Holz (2008) A simple semi-analytical model for remote estimation
of chlorophyll-a in turbid productive waters: Validation. Remote Sensing of
Environment, 112, 3582-3593.
Gitelson, A.-A., Gritz, Y., M.-N. Merzlyak (2003) Relationships between leaf
chlorophyll content and spectral reflectance and algorithms for non-destructive
chlorophyll assessment in higher plant leaves. Journal of Plant Physiology, 160, 271–
282.
Gitelson, A. (1993) The nature of the peak near 700 nm on the radiance spectra and its
application for remote estimation of phytoplankton pigments in inland waters.
Proceedings SPIE, 8th Meeting on Optical Engineering in Israel: Optical
Engineering and Remote Sensing, 1971, 170-179.
Gitelson, A., Szilagyi, F., and K.-H. Mittenzwey (1993) Improving quantitative remote
sensing for monitoring of inland water quality. Water Research, 27, 1185-1194.
Gitelson, A. (1992) The peak near 700 nm on radiance spectra of algae and water:
relationships of its magnitude and position with chlorophyll concentration.
International Journal of Remote Sensing, 13, 3367-3373.
Gitelson, A.-A. and K.-Y. Kondratyev (1991) Mechanism producing the near-700 nm
brightness peak in the emission spectra of water bodies and its possible applications
in remote sounding of such bodies. Transactions of the U.S.S.R. Academy of
Sciences: Earth Science Sections, 306, 1-4.
Gitelson, A.-A., Nikanorov, A.-M., Szabo, G.-Y., and F. Szilagyi (1986) Etude de la
qualité des eaux de surface par télédétéction. Monitoring to Detect Changes in Water
Quality Series (Proceedings of the Budapest Symposium, July 1986), IAHS Publ. no.
157, 111-121.
Gitelson, A., Keydan, G., and Shishkin, V. (1985) Inland waters quality assessment from
satellite data in visible range of the spectrum. Sovjet Journal of Remote Sensing, 6,
28-36.
Glazer, A.-N., Fang, S., and D.-M. Brown (1973) Spectroscopic properties of Cphycocyanin and of its α and β subunits. The Journal of Biological Chemistry, 248,
5679-5685.
125
Goldman, C.-R. (1988) Primary productivity, nutrients, and transparency during the early
onset of eutrophication in ultra-oligotrophic Lake Tahoe, California-Nevada,
Limnology and Oceanography, 33, 1321-1333.
Goldman, C.-R. (1968) Aquatic Primary Production. American Zoologist, 8, 31-42.
Gons, H.-J., Auer, M.-T., and S.-W. Effler (2008) MERIS satellite chlorophyll mapping
of oligotrophic and eutrophic waters in the Laurentian Great Lakes. Remote Sensing
of Environment, 112, 4098-4106.
Gons, H.-J., Rijkeboer, M., & Ruddick, K.-G. (2005) Effect of a waveband shift on
chlorophyll retrieval from MERIS imagery of inland and coastal waters. Journal of
Plankton Research, 27, 125-127.
Gons, H.-J., Rijkeboer, M., and K.-G. Ruddick (2002) A chlorophyll-retrieval algorithm
for satellite imagery (Medium Resolution Imaging Spectrometer) of inland and
coastal waters, Journal of Plankton Research, 24, 947-951.
Gons, H.-J., Rijkeboer, M., Bagheri, S., and K.-G. Ruddick (2000) Optical teledetection
of chlorophyll a in estuarine and coastal waters. Environmental Science and
Technology, 34, 5189-5192.
Gons, H.-J. (1999) Optical teledetection of chlorophyll a in turbid inland waters.
Environmental Science and Technology, 33, 1127-1132.
Gordon, H.-R., Brown, O.-B., Evans, R.-H., Brown, J.-W., Smith, R.-C., Baker, K.-S.,
and D.-K. Clark (1988) A semianalytic radiance model of ocean color, Journal of
Geophysical Research, 93(D9), 10909-10924.
Gordon, H.-R., Clark, D.-K., Brown, J.-W., Brown, O.-B., Evans, R.-H., and W.-W.
Broenkow (1983) Phytoplankton pigment concentrations in the Middle Atlantic
Bight: comparison between ship determinations and Coastal Zone Color Scanner
estimates, Applied Optics, 22, 20-36.
Gordon, H.-R. and A.-Y. Morel (1983) Remote assessment of ocean color for
interpretation of satellite visible imagery – a review. In: Barber, R.-T., Mooers, C.-N.K., Bowman, M.-J. and B. Zeitzschel, ed. Lecture Notes on Coastal and Estuarine
Studies 4. New York: Springer Verlag.
Gordon, H.-R., Brown, O.-B., and M.-M. Jacobs (1975) Computed relationships between
the inherent and apparent optical properties of a flat homogeneous ocean. Applied
Optics, 14, 417-427.
Gordon, H.-R. (1973) Simple calculation of the diffuse reflectance of the ocean. Applied
Optics, 12, 2803-2804.
126
Gower, J. and S. King (2007) Validation of chlorophyll fluorescence derived from
MERIS on the west coast of Canada. International Journal of Remote Sensing, 28,
625 – 635.
Gower, J., King, S., Borstad, G., and L. Brown (2005) Detection of intense plankton
blooms using the 709 nm band of the MERIS imaging spectrometer. International
Journal of Remote Sensing, 26, 2005-2012
Gower, J.-F.-R., Doerffer, R., and G.-A. Borstad (1999) Interpretation of the 685nm peak
in water-leaving radiance spectra in terms of fluorescence, absorption and scattering,
and its observation by MERIS. International Journal of Remote Sensing, 20, 17711786.
Gower, J.-F.-R. and Borstad (1990) Mapping of phytoplankton by solar-stimulated
fluorescence using an imaging spectrometer. International Journal Remote Sensing,
11, 313-320.
Gower, J.-F.-R. (1980) Observations of in situ fluorescence of chlorophyll-a in Saanich
inlet. Boundary-Layer Meteorology, 18, 235-245.
Grossmann, B.-E. & Browell, E.-V. (1989) Spectroscopy of water vapour in the 720 nm
wavelength region: Line strengths, self-induced pressure broadenings and shifts, and
temperature dependence of linewidths and shifts. Journal of Molecular Spectroscopy,
136, 262-294.
Gurlin, D., Gitelson, A.A., and W.J. Moses (2011) Remote estimation of chl-a
concentration in turbid productive waters - Return to a simple two-band NIR-red
model? Remote Sensing of Environment, 115, 3479-3490.
Han, L. (1997) Spectral reflectance with varying suspended sediment concentrations in
clear and algae-laden waters. Photogrammetric Engineering and Remote Sensing, 63,
701-705.
Han, L. and D.-C. Rundquist (1997) Comparison of NIR/RED ratio and first derivative of
reflectance in estimating algal-chlorophyll concentration: a case study in a turbid
reservoir. Remote Sensing of Environment, 62, 253-261.
Hayase, K. and H. Tsubota (1985) Sedimentary humic acid and fulvic acid as fluorescent
organic materials. Geochimica et Cosmochimica Acta, 49, 159-163.
Hunter, P.-D., Tyler, A.-N., Carvalho, L., Codd, G.-A., and S.-C. Maberly (2010)
Hyperspectral remote sensing of cyanobacterial pigments as indicators for cell
populations and toxins in eutrophic lakes. Remote Sensing of Environment, 114,
2705-2718.
127
Hunter, P.-D., Tyler, A.-N., Gilvear, D.-J., and N.-J. Willby (2009) Using remote sensing
to aid the assessment of human health risks from blooms of potentially toxic
cyanobacteria. Environmental Science and Technology, 43, 2627-2633.
Kallio, K., Kutser, T., Hannonen, T., Koponen, S., Pulliainen, J., Vepsäläinen, J., and T.
Pyhälahti (2001) Retrieval of water quality parameters of various lake types in
different seasons. Science of the Total Environment, 268, 59-77.
Kowalczuk, P., Olszewski, J., Darecki, M., and S. Kaczmarek (2005) Empirical
relationships between coloured dissolved organic matter (CDOM) absorption and
apparent optical properties in Baltic Sea waters. International Journal of Remote
Sensing, 26, 345-370.
Kutser, T., Hiire, M., Metsamaa, L., Vahtmäe, E., Paavel, B., and R. Aps (2009) Field
measurements of spectral backscattering coefficient of the Baltic Sea and boreal
lakes. Boreal Environment Research, 14, 305-312.
Koponen, S., Attila, J., Pulliainen, J., Kallio, K., Pyhälahti, T., Lindfors, A., Rasmus. K.,
and M. Hallikainen (2007) A case study of airborne and satellite remote sensing of a
spring bloom event in the Gulf of Finland. Continental Shelf Research, 27, 228-244.
Koponen, S., Pulliainen, J., Kallio, K., and M. Hallikainen (2002) Lake water clarity
classification with airborne hyperspectral spectrometer and simulated MERIS data.
Remote Sensing of Environment, 79, 51-59.
Kou, L., Labrie, D., and Chylek, P. (1993) Refractive index of water and ice in the 0.65to 2.5-µm spectral range. Applied Optics, 32, 3531-354.
Le, C., Li, Y., Zha, Y., Sun, D., Huang, C., and H. Lu (2009) A four band semi-analytical
model for estimating chlorophyll-a in highly turbid lakes: The case ofTaihu Lake,
China. Remote Sensing of Environment, 113, 1175-1182.
Li, L., Sengpiel, R.-E., Pascual, D.-L., Tedesco, L.-P., Wilson, J.-S. and E. Soyeux
(2010) Using hyperspectral remote sensing to estimate chlorophyll-a and
phycocyanin in a mesotrophic reservoir. International Journal of Remote Sensing, 31,
4147-4162.
Lillesand, T.-M., Johnson, W.-L., Deuell, R.-L., Lindstorm, O.-M., and D.-E. Meisner
(1983) Use of Landsat data to predict the trophic state of Minnesota Lakes.
Photogrammetric Engineering and Remote Sensing, 49, 219-229
Mitchell, B.-G., Kahru, M., Wieland, J., and Stramska, M. (2003) Determination of
spectral absorption coefficients of particles, dissolved material and phytoplankton for
discrete water samples. In: J.-L. Mueller, G.-S. Fargion, and C.-R. McClain (Eds.)
Inherent optical properties: Instruments, characterizations, field measurements and
data analysis protocols. Ocean Optics Protocols for Satellite Ocean Color Sensor
128
Validation, revision 4, volume IV, Goddard Space Flight Center Technical
Memorandum 2003-211621.
Mitchell, B.-G. (1990) Algorithms for determining the absorption coefficient for aquatic
particulates using the quantitative filter technique. Proceedings SPIE, 1302, 137-148.
Mittenzwey, K.-H., Ullrich, S., Gitelson, A.-A., and K.-Y. Kondratiev (1992)
Determination of chlorophyll a of inland waters on the basis of spectral reflectance.
Limnolgie and Oceanography, 37, 147-149.
Mittenzwey, K.-H., Breitwieser, S., Penig, J., Gitelson, A.-A., Dubovitzkii, G., Garbusov,
G., Ullrich, S., Vobach, V., and A. Müller (1991) Fluorescence and reflectance for the
in-situ determination of some quality parameters of surface waters. Acta
hydrochimica et hydrobiologica, 19, 3-15.
Morel, A. and B. Gentili (1993) Diffuse reflectance of oceanic waters. II. Biderectional
aspects., Applied Optics, 32, 6864-6879
Morel, A. and B. Gentili (1991) Diffuse reflectance of oceanic waters: its dependence on
Sun angle as influenced by the molecular scattering contribution. Applied Optics, 30,
4427-4438.
Morel, A. and L. Prieur (1977) Analysis of variations in ocean color. Limnology and
Oceanography, 22, 709-722.
Moses, W.-J., Gitelson, A.-A., Perk, R.-L., Gurlin, D., Rundquist, D.-C., Leavitt, B.-C.,
Barrow, T.-M., and P. Brakhage (2012) Estimation of chl-a concentration in turbid
productive waters using airborne hyperspectral data. Water Research, 993-1004
Moses, W. J., A.A. Gitelson, S. Berdnikov, V. Saprygin, V. Povazhnyi (2012)
Operational MERIS-based NIR-red algorithms for estimating chlorophyll-a
concentrations in coastal waters - The Azov Sea case study, Remote Sensing of
Environment, 121: 118–124.
Moses, W.-J., Gitelson, A.-A., Berdnikov, S., and V. Povazhnyy (2009a) Estimation of
chlorophyll-a concentration in case II waters using MODIS and MERIS datasuccesses and challenges. Environmental Research Letters, 4, 045005, 8pp.
Moses, W.-J, Gitelson, A.-A., Berdnikov, S., V. Povazhnyy (2009b) Satellite Estimation
of Chlorophyll-a Concentration Using the Red and NIR Bands of MERIS - The Azov
Sea Case Study. IEEE Geoscience and Remote Sensing Letters, 6, 845-849.
Mueller, J.-L. (2003) Inherent optical properties: instrument characterizations, field
measurements and data analysis protocols. In: Ocean Optics Protocols for Satellite
Ocean Color Sensor Validation, Revision 4, Volume IV, Erratum 1 to Pegau, S.,
Zanefeld, J.-R.-V., and Mueller, J.-L. (2003) Inherent optical property measurement
129
concepts: physical principles and instruments. In: J.-L. Mueller, G.-S. Fargion, and
C.-R. McClain (Eds.) Inherent optical properties: Instruments, characterizations,
field measurements and data analysis protocols. Ocean Optics Protocols for Satellite
Ocean Color Sensor Validation, revision 4, volume IV, Goddard Space Flight Center
Technical Memorandum 2003-211621.
NDEQ Water Quality Division (2010) 2010 Water Quality Integrated Report. NDEQ
Integrated Reports, Nebraska Department of Environmental Quality, Nebraska, USA.
Neville, R.-A. and Gower, J.-F.-R. (1977) Passive remote sensing of phytoplankton via
chlorophyll a fluorescence. Journal of Geophysical Research, 82, 3487-3493.
Nguy-Robertson, A.-L., Li, L., Tedesco, L., Wilson, J., and E. Soyeux (2012) Comparing
the performance of empirical, semi-empirical, and curve fitting models in predicting
cyanobacterial pigments. Photogrammetric Engineering & Remote Sensing, 78, 485494.
Nusch, E.-A. (1980) Comparison of different methods for chlorophyll and phaeopigment
determination. Archiv für Hydrobiologie–Beiheft Ergebnisse der Limnologie, 14, 1436.
Ohde, T. and Siegel, H. (2003) Derivation of immersion factors for the hyperspectral
TriOS radiance sensor. Journal of Optics A – Pure and Applied Optics, 5, L12-L14.
O’Reilly, J.-E., Maritorena, S., O’Brien, M.-C., Siegel, D.-A., Toole, D., Menzies, D.,
Smith, R.-C., Mueller, J.-L., Mitchell, B.-G., Kahru, M., Chavez, F.-P., Strutton, P.,
Cota, G.-F., Hooker, S.-B., McClain, C.-R., Carder, K.-L., Müller-Karger, F.,
Harding, L., Magnuson, A., Phinney, D., Moore, G.-F., Aiken, J., Arrigo, K.-R.,
Letelier, R., and Culver, M. (2000) SeaWiFS Postlaunch Calibration and Validation
Analyses. In: S.-B. Hooker, and E.-R. Firestone (Eds.) SeaWiFS Postlaunch
Technical Report Series, 11 (3), NASA Technical Memorandum 2000–206892,
Volume 11.
O'Reilly, J. E., Maritorena, S., Mitchell, B. G., Siegel, D. A., Carder, K. L., Garver, S. A.,
Kahru, M. and McClain, C. (1998) Ocean color chlorophyll algorithms for SeaWiFS.
Journal of Geophysical Research-Oceans, 103(C11), 24937-24953.
Pegau, W.-S., Gray, D., and J.-R.V. Zaneveld (1997) Absorption and attenuation of
visible and near-infrared light in water: dependence on temperature and salinity.
Applied Optics, 36, 6035-6046.
Pope, R.-M. and Fry, E.-S. (1997) Absorption spectrum (380 – 700 nm) of pure water. II.
Integrating cavity measurements. Applied Optics, 36, 8710-8723.
Preisendorfer, R.-W. (1986) Secchi disk science: Visual optics of natural waters.
Limnology and Oceanography, 31, 909-926
130
Prieur, L. and S. Sathyendranath (1981) An optical classification of coastal and oceanic
waters based on the specific spectral absorption curves of phytoplankton pigments,
dissolved organic matter, and other particulate materials. Limnology and
Oceanography, 26, 671-689.
Quibell, G. (1991) The effect of suspended sediment on reflectance from freshwater
algae. International Journal of Remote Sensing, 12, 177-182.
Randolph, K., Wilson, J., Tedesco, L., Li, L., Pascual, D.-L., and E. Soyeux (2008)
Hyperspectral remote sensing of cyanobacteria in turbid productive waters using
optically active pgments, chlorophyll-a and phycocyanin. Remote Sensing of
Environment, 112, 4009-4019.
Revenga, C. and Kura, Y. (2003) Status and trends of biodiversity of inland ecosystems.
Secretariat of the Convention on Biological Diversity, Montreal, Technical Series no.
11 116 pages.
Reynolds, C.-S. (1998) What factors influence the species composition of phytoplankton
in lakes of different trophic status? Hydrobiologica, 369/370, 11-26
Reynolds, C.-S. (1987) The response of phytoplankton communities to changing light
environments. Aquatic Sciences, 49, 220-236.
Ritchie, R.-J. (2008) Universal chlorophyll equations for estimating chlorophylls a, b, c,
and d and total chlorophylls in natural assemblages of photosynthetic organisms using
acetone, methanol, or ethanol solvents. Photosynthetica, 46, 115-126.
Ruiz-Verdú, A.-R., Simis, S.-G.-H., de Hoyos, C., Gons, H.-J., Peña-Martínez, R. (2008),
Ryan, J.-P., Fischer, A.-M., Kudela, R.-M., Gower, J.-F.-R., King, S.-A., Marin III, R.,
and F.-P. Chavez (2009) Influences of upwelling and downwelling winds on red tide
bloom dynamics in Monterey Bay, California. Continental Shelf Research, 29, 785795.
Sarada, R., Pillai, M.-G., and G.-A. Ravishankar (1999) Phycocyanin from Spirulina sp:
influence of processing of biomass on phycocyanin yield, analysis of efficacy of
extraction methods and stability studies on phycocyanin. Process Biochemistry, 34,
795-801.
Schalles, J.-F. and Y.-Z. Yacobi (2000) Remote detection and seasonal patterns of
phycocyanin, carotenoid and chlorophyll pigments in eutrophic waters. Archiv fuer
Hydrobiologie Special Issues, Advances in Limnology, Limnology and Lake
Management 2000+, 55, 153-168.
Schindler, D.-W. (2006) Recent advances in the understanding and management of
eutrophication. Limnology and Oceanography, 51, 356-363.
131
Schindler, D.-W. (1978) Factors regulating phytoplankton production and standing crop
in the world’s freshwaters. Limnology and Oceanography, 23, 478 – 486.
Shumway, S.-E. (1990) A review of the effects of algal blooms on shellfish and
aquaculture. Journal of the World Aquaculture Society, 21, 65-104
Simis, S.-G.-H., Ruiz-Verdú, A., Domínguez-Gómez, J.-A., Peña-Martinez, R., Peters,
S.-W.-M., and H. Gons (2007) Influence of phytoplankton pigment composition on
remote sensing of cyanobacterial biomass. Remote Sensing of Environment, 106, 414427.
Simis, S.-G.-H., Peters, S.-W.-M. and Gons, H.-J. (2005) Remote sensing of the
cyanobacterial pigment phycocyanin in turbid inland water. Limnology and
Oceanography, 50, 237-245.
Smith, R.-C. and W.-H. Wilson (1981) Ship and satellite bio-optical research in the
California Bight. In: J.-R.-F. Gower, ed. Oceanography from Space, Plenum Press,
New York, 281-294.
Spectral Imaging Ltd. (2012) AisaEAGLE hyperspectral sensor, AisaEAGLE data sheet,
Oulu, Finland: Spectral Imaging Ltd.
Strand, H., Hoeft, R., Stritthold, J., Miles, L., Horning, N., Fosnight, E., and W. Turner
(Eds.), (2007) Sourcebook on remote sensing and biodiversity indicators. Secretariat
of the Convention on Biological Diversity, Montreal, Technical Series no. 32, 203
pages
Strömbeck, N. and D.-C. Pierson (2001) The effects of variability in the inherent optical
properties on estimations of chlorophyll a by remote sensing in Swedish reservoirs.
The Science of the Total Environment, 268, 123-137
Stumpf, R.-P. & Tyler, M.-A. (1988) Satellite detection of bloom and pigment
distributions in estuaries. Remote Sensing of Environment, 24, 385-404.
Sun, D., Li, Y., Wang, Q., Le, C., Huang, C., and K. Shi (2011) Development of optical
criteria to discriminate various types of highly turbid waters. Hydrobiologia, 669, 83104.
Thiemann, S. and H. Kaufmann (2002) Lake water quality monitoring using
hyperspectral airborne data – a semi-empirical multisensory and multitemporal and
multispectral approach for the Mecklenburg Lake District, Germany. Remote Sensing
of Environment, 81, 228-237.
132
Thomas, E.-A. (1969) The concept of eutrophication in central European Lakes. In
Eutrophication: causes, consequences, correctives, Proceedings of a Symposium,
National Academy of Sciences, Washington, DC, 29-49.
Tyler, A.-N., Hunter, P.-D., Carvalho, L., Codd, G.-A., Elliot, J.-A., Ferguson, C.-A.,
Hanley, N.-D., Hopkins, D.-W., Maberly, S.-C., Mearns, K.-J., and E.-M. Scott
(2009) Strategies for monitoring and managing mass populations of toxic
cyanobacteria in recreational waters: a multi-disciplinary approach. Environmental
Health, 8, 1-8.
United Nations Conference on Environment and Development (1992) Protection of the
quality & supply of freshwater resources: application of integrated approaches to the
development, management & use of water resources. Agenda 21: Programme of
Action for Sustainable Development, chapter 18, Rio de Janeiro, 3 to 14 June 1992,
UN Doc A/Conf.151/26 (1992).
Vasilkov, A. and O. Kopelevich (1982) Reasons for the appearance of the maximum near
700 nm in the radiance spectrum emitted by the ocean layer. Oceanology, 22, 697701.
Vollenweider, R.-A. (1968) Scientific fundamentals of the eutrophication in lakes and
flowing waters, with particular reference to nitrogen and phosphorus as factors in
eutrophication. Organisation for Economic Co-operation and Development, Paris,
Technical Report DAS/CSI/68.27.
Welschmeyer, N.-A. (1994) Fluorometric analysis of chlorophyll a in the presence of
chlorophyll b and phaeopigments. Limnology and Oceanography, 39, 1985-1992.
WET Labs, Inc. (2008) ECO 3-measurement sensor (triplet) user’s guide. (revision O),
Philomath: WET Labs, Inc.
World Water Assessment Programme (2009) The United Nations World Water
Development Report 3: Water in a Changing World. Paris: UNESCO Publishing, and
London: Earthscan.
World Water Assessmenr Programme (2006) The United Nations World Water
Development Report: Water – a shared responsibility. Paris: UNESCO Publishing,
and Barcelona: Berghahn Books.
World Water Assessment Programme (2003) The United Nations World Water
Development Report: Water for People - Water for Life. Paris: UNESCO Publishing,
and Barcelona: Berghahn Books.
Yacobi, Y.-Z., Moses, W.-J., Kaganovsky, S., Sulimani, B., Leavitt, B.-C., and A.-A.
Gitelson (2011) NIR-red reflectance-based models for chlorophyll-a estimation in
mesotrophic inland and coastal waters: Lake Kinneret case study. Water Research,
45, 2428-2436.
133
Yacobi, Y.-Z., Gitelson, A.-A.,, and M. Mayo (1995) Remote Sensing of chlorophyll in
Lake Kinneret using high spectral resolution radiometer and Landsat TM: Spectral
features of reflectance and algorithm development. Journal of Plankton Research, 17,
2155-2173.
Yang, W., Matsushita, B., Chen, J., Fukushima, T., and R, Ma (2010) An enhanced threeband index for estimating chlorophyll-a in turbid case-II waters: case studies of Lake
Kasumigaura, Japan, and Lake Dianchi, China. IEEE Transactions on Geoscience and
Remote Sensing, 7, 655-659.