Retrieval of ocean properties
using multispectral methods
S. Ahmed, A. Gilerson, B. Gross, F. Moshary
Students: J. Zhou, M. Vargas, A. Gill, B. Elmaanaoui, K. Aran
Spectral Algorithm Development for Sensing of Coastal Waters
Separation of Overlapping Elastic Scattering and Fluorescence
from Algae in Seawater through Polarization Discrimination
1
Spectral Algorithm Development for
Sensing of Coastal Waters
Reflectance curves from the 2002 cruise in Peconic Bay, Long Island
2
Ratio algorithm performance –
Eastern Long Island
Blue / Green
NIR Spectral Ratio
200
0.4
690/670 = 2.0898*Chl + 99.549
2
180
R = 0.8726
160
690/670
R440/R550
0.3
0.2
y = 0.3256e-0.0217x
0.1
140
120
2
R = 0.7879
100
0
0
5
10
15
20
25
Chlorophyll, mg/m3
30
35
80
0
10
20
30
40
Chlorophyll-a, mg/m3
In homogeneous waters where only Chlorophyll varies Blue / Green works only in
Case I (see later) NIR Ratios work well in both Case I and Case II
3
but may be limited by small signals in open waters
Absorption/Backscatter features
1
2 3
1- Chlorophyll absorption can be probed effectively using 440-570 band ratios
2- In presence of TSS and CDOM, Blue-Green ratios are contaminated.
3- Red-NIR algorithms are much less sensitive to TSS, CDOM.
4
4- The 670-710 channels effectively probe the ChL absorption feature and the
730 channel effectively calculates the backscatter since water abs dominates
Simulation
Blue-Green
Three Band NIR ratios
Very high spread in the Blue-Green Ratio due to CDOM and TSS randomized
variability. This aspect is not relevant to the Red/NIR algorithms
5
Multispectral versus Hyperspectral
assessment of GOES-R Coastal Water Imager
• Future sensors (GOES-R) need to decide
between multispectral or hyperspectral mode.
• Hyperspectral channels are very important for
shallow water retrieval
• Preliminary tests compared multispectral vs
hyperspectral sensing schemes based on
Hydrolight Radiative transfer derived biooptical model.
6
Shallow Water Bio-Optical Model
Based on Hydrolight RT simulations
(Carder et al)
Parameterized Shallow Water Model Parameters
P
Phytoplankton Absorption at 440nm
Deep
Shallow
G
Gelbstoff Absorption at 440nm
Deep
Shallow
X
Backscatter Amplitude at 440 nm
Deep
Shallow
Y
Backscatter Power Exponent
Deep
Shallow
H
Ocean Column Depth
Shallow
B
Bottom Surface Albedo
Shallow
7
Remote Sensing Reflectance Spectra
Inversion error versus measurement noise
for all 6 parameters
Normalized Parameter Retrieval Error
Bottom Albedo
Phytoplankton
0.9
Gelbstoff
0.9
6p hyperspectral
6p multispectral
0.8
3.5
6p hyperspectral
6p multispectral
0.8
6p hyperspectral
6p multispectral
3
0.7
0.7
0.6
0.6
0.5
0.5
2
0.4
0.4
1.5
0.3
0.3
0.2
0.2
0.1
0.1
2.5
1
0
0
2
4
6
8
10
12
14
16
18
20
0
0.5
0
2
4
6
Phytoplankton
8
10
12
14
16
18
20
0
2
4
Height
0.9
6p hyperspectral
6p multispectral
8
10
12
14
16
18
20
16
18
20
0.2
6p hyperspectral
6p multispectral
0.2
0.7
6
Power Exponent
0.25
0.8
0
6p hyperspectral
6p multispectral
0.18
0.16
0.14
0.6
0.15
0.12
0.5
0.1
0.4
0.1
0.08
0.3
0.06
0.2
0.05
0.04
0.1
0
0.02
0
2
4
6
8
10
12
14
16
18
20
0
0
2
4
6
8
10
12
14
Noise (%)
16
18
20
0
0
2
4
6
8
10
12
14
8
Results
• Hyperspectral channels are absolutely needed
to reduce errors in shallow bottom heights and
bottom reflectance (Panels 1 and 5)
• Ocean column parameters are also much better
retrieved using Hyperspectal configuration
except for spectral slope of backscatter
parameter which makes sense since this
parameter caused only broad modification of
the reflectance spectra. (Panel 6)
9
• Chl retrieval in Productive Case I waters can be obtained by
both conventional blue-green type algorithms as well as NIR
ratio algorithms
• TSS and CDOM variability in case II waters makes blue/green
ratios useless but three band NIR ratios are very insensitive to
these parameters
• Ratio algorithms for case II waters need thorough testing with
in-situ monitoring using a consistent field testing protocol.
• The effects of atmospheric correction to assess the sensitivity
of the various two and three ratio algorithms need to be
explored.
• Development and sensitivity analysis of simultaneous
atmosphere /ocean parameter retrieval using both multispectral
and hyperspectral algorithms
10
Separation of Overlapping Elastic Scattering
and Fluorescence from Algae in Seawater
through Polarization Discrimination
Objective: Separate overlapping fluorescence and elastic
scattering spectra of algae excited by white light
Method: Utilize polarization properties of elastically scattered
light and unpolarized nature of excited fluorescence to
separate the two
Applications: Use fluorescence obtained as indication of Chl
concentration even in turbid waters
Obtain elastic scattering spectra free of overlapping
fluorescence for ocean color work
11
Reflectance curves from the 2002
cruise in Peconic Bay, Long Island
12
Fluorescence Height
Reflectance
Fluorescence Height
670
685
745
Wavelength, nm
Traditional method of the fluorescence height calculation over baseline
13
0.05
Fluorescence height
over baseline
Fluorescence
Reflectance
0.04
Reflectance +
fluorescence
Reflectance
0.03
0.02
Reflectance peak
at minimum absorption
665nm
0.01
600
650
685nm
700
746nm
750
800
Wavelength, nm
14
Experimental Setup
FP
Illuminator
i2
Spectrometer
i1
P2
L
Nozzle
P1
WL
θ
C
L – lens, FP – fiber probe, A – aperture, P1, P2 – polarizers,
C – cuvette with algae, WL – water level.
Objects tested: algae Isochrysis sp., Tetraselmis striata,
Thalassiosira weissflogii, “Pavlova”, concentrations up to 4x10^6 cells/mL,
15
algae with clays.
Polarized Illumination
1.2
Reflectance, a.u.
1.0
Rmax ( )  R ( )  0.5Fl ( ),
Rmax )
Rmin ( )  R| | ( )  0.5 Fl ( ),
0.8
Near zero if no depolarization
valid for spherical particles
0.6
0.4
Rmin()
Fllaser
0.2
Fl ( )  2Rmin ( )
Rmax ( )  Rmin ( )  R ( )  0.5Fl ( )
0.0
500
600
Wavelength, nm
700
 R|| ( )  0.5Fl ( )  R ( )  R|| ( )
Generally validated using laser induced fluorescence but significant
error results due to scattering component
16
Extracted Fluorescence
1.2
0.14
RD
0.12
R
0.8
0.6
Fllaser 
0.4
Fl 
0.2
Reflectance, a.u.
1.0
Reflectance, a.u.
R
Rs 
0.10
0.08
Rs 
0.06
RD
0.04
Fl 
0.02
0.0
0.00
500
600
700
500
Wavelength, nm
600
700
Wavelength, nm
Algae Isochrysis sp.
Algae Tetraselmis striata
(brown algae spherical d ≈ 5 µm)
(green algae slightly ellipsoidical d ≈ 12 µm)
R( )  Rmax    Rmin   ;
RD ( )  R    R||   ;
Technique with polarized light
Rs ( )  A * RD    B
17
Unpolarized source
Light scattered by the algae illuminated by unpolarized light has some degree
of polarization and can be also analyzed using polarization discrimination with
the same linear regression approach
1.0
0.8
Rs 
0.6
Rmax 
0.7
Rmax
R(
0.6
Fl 
0.4
Rmin
0.2
Reflectance
Reflectance, a.u.
0.8
RmaxFl
0.5
0.4
Rmin
0.3
RminFl
0.2
0.1
0.0
500
600
700
Wavelength, nm
0.0
400
500
600
700
Wavelength, nm
Algae Isochrysis sp. (brown algae
spherical d ≈ 5 µm)
18
Algae with clay
0.016
0
10
50
100
200
Reflectance
0.012
0.010
Magnitude of fluorescence
0.014
Clay conc
Cs, mg/l
0.008
0.006
0.004
0.002
0.000
500
600
700
Wavelength, nm
Reflectance curves for algae with
clay, Cs = 0 - 200 mg/l
0.0034
0.0032
0.0030
0.0028
0.0026
0.0024
0.0022
0.0020
0.0018
0.0016
0.0014
0.0012
0.0010
0.0008
0.0006
0.0004
0.0002
0.0000
unpolarized light
polarized light
50
100
150
200
Clay concentration, mg/l
Fluorescence magnitude retrieved from
algae with different concentrations of
clay
Clay – Na-Montmorillonite, particle size 2-4 µm
19
Extraction of fluorescence in the waters
with rough surface (lab experiments)
Unpolarized light
0.20
0.30
0.25
R
R
0.10
Fl 
Rs 
0.05
RD 
Reflectance, a.u.
Reflectance, a.u.
0.15
0.20
Rs 
0.15
RD 
0.10
0.05
Fl 
0.00
0.00
-0.05
500
600
700
Wavelength, nm
500
600
700
Wavelength, nm
Probe above the water, probe vertical
No wind
Wind speed above the surface ≈ 9.5 m/s
Sample time increased to 10s from 1s
Algae Isochrysis. Concentration ~4.0 mln cells/ml.
20
Extraction of fluorescence
in the waters of Shinnecock Bay,
Long Island
0.10
1.2
Rs 
0.08
0.06
Rmax 
R(
Perp/Par
Reflectance, a.u.
1.0
0.04
Rmin
RD 
0.02
Ratio of perp and par components
Boat Hampton Bay 060904
0.8
0.6
0.4
0.00
0.2
Fl 
0.0
-0.02
500
600
700
Wavelength, nm
Chl concentration about 8 µg/l
June 2004
400
500
600
700
Wavelengths, nm
Ratio between 2
polarization components is
close to linear
21
800
Simulation Model for Case 2
Waters
Input
bb ( )  N pl b , pl ( )  N min  b ,min ( )  bbw ( )
a( )  a w ( )  a pl ( )  N min  a ,min ( )  a y ( )
a pl ( )  0.06ac ( )C 0.65
*
- Backscattering coefficient
- Absorption coefficient
[Mobley, 1994]
- Absorption coefficient of phytoplankton
[Morel, 1991]
a y ( )  a y (0 ) exp[ 0.014(  0 )]
- Absorption coefficient of CDOM
amin ( )  amin (0 ) exp[ 0.009 ]
- Absorption coefficient of minerals
R ( )  0.33bb (  /( a(   bb ( )
675
EFl    ( Ed ( )a pl ( ) / a( )) d
400
[Bricaud, et al., 1981]
[Stramski, et al., 2001]
- Reflectance
[Morel, 1977]
- Energy of emitted fluorescence
[Gower, et al., 1999]
22
Simulation model for case 2
waters
Output
Polarization components of reflectance are calculated from Mie
code for 45° illumination (30° in water) & vertical observation
R (   0.33 * 2S150 ( ) /( a( )  2S150 ( ))
R|| (   0.33 * 2S150|| ( ) /( a( )  2S150 ( ))
S150 ( )
-scattering function at 150°, which was used as average value for calculating
backscattering
Polarization components of S150 ( ) were used for calculation of reflectance polarization components
where

Half of fluorescence is superimposed on polarization components
as a spectrum with Gaussian shape centered at 685 nm
Rmax ( )  R ( )  0.5 Fl ( ),
Rmin ( )  R|| ( )  0.5 Fl ( )
Fluorescence is retrieved using polarization technique
Fl    2( ARmin ( )  B  Rmax ( )) /( A  1)
A and B are determined from fitting
outside fluorescence zone
Rmax ( )  ARmin ( )  B
23
Simulation Model Results
0.030
0.10
Cs = 100 mg/l
b
a
0.025
0.08
3
C=5mg/m , Cs=10mg/l
C = 50 mg/m
Reflectance
Reflectance
0.020
0.015
0.010
0.005
3
0.06
Cs = 40 mg/l
0.04
0.02
Cs = 10 mg/l
0.000
0.00
400
500
600
Wavelength, nm
700
800
400
500
600
700
800
Wavelength, nm
Fluorescence retrieval from reflectance spectra for different
concentrations of mineral particles: a) C = 5 mg/m3, b) C = 50
mg/m3.
24
Results of fluorescence retrieval,
comparison with baseline method
0.05
0.022
Fluorescence height
over baseline
0.020
Magnitude of fluorescence
Fluorescence
Reflectance
0.04
Reflectance +
fluorescence
Reflectance
0.03
0.02
Reflectance peak
at minimum absorption
0.018
Fl height
0.016
0.014
0.012
Fltheor
0.010
Flretr
0.008
0.006
665nm
0.01
600
685nm
746nm
0.004
650
700
Wavelength, nm
750
800
0
50
100
150
200
Concentration of particles, mg/l
Comparison of retrieved fluorescence peak to assumed values
for a range of mineral particle concentrations using both
polarization discrimination and baseline subtraction
25
Conclusions/Future Work
• Separation of Chlorophyll Fluorescence from scattering using
polarization discrimination has been demonstrated for 4 types
of algae with different shapes, sizes of particles
• Implementation of the technique using both white light and
sun light sources has proven successful in the lab and in the
field conditions
• Fluorescence extraction has been obtained even with the
presence of high concentration of scattering medium
• Validation with laser induced fluorescence has been performed
• Extraction of fluorescence is successful for all illumination
angles with polarized light, up to 50 deg for unpolarized light.
26
Conclusions/Future Work
• Magnitude of fluorescence peak extracted from reflectance
spectra through polarization technique does not change with
the concentration of scattering medium up to 200 mg/l.
• Computer simulations show that fluorescence can be
successfully retrieved for most water conditions typical for
coastal zones with accuracy 7-11%.
• “Fluorescence height” over baseline strongly overestimates
actual and retrieved fluorescence height and these values do
not correlate with each other for different concentrations of
mineral particles.
• Future simulations should include effects of multiple scattering
and atmosphere on polarization components and fluorescence
retrieval process.
27
Long Island Field Measurements
28
Bio-Optical Model 1
0.5rrs
RRS   
1  1.5rrs
RRS  Above water rrs  Below water
rrs  rrsc  rrsB
Due to column and water floor respectively

  1
 
c
dp 
c
rrs  rrs 1  exp  
 Du H  




cos



w






  1

1
B
B
rrs   B exp  
 Du H 

  cos w 
 
rrsdp  0.084  .170u u
bb
u
a  bb
Duc  1.03 1  2.4u 0.5
DuB  1.04 1  5.4u 0.5
  a  bb
bb    total backscatte r a   total extinction
29
Bio-Optical Model 2
atotal    aw    a    ag   m1
aw  
is the absorption coefficient due to water
a  
is the absorption coefficient due to phytoplankton
a g  
is the absorption coefficient due to gelbstoff
btot    bbw ( )  bbp ( )
bbw ( )
is the backscattering of water
bbp ( ) is the backscattering by particulate matters
30
Bio-Optical Model 3
a g    G exp(  S (  400))
S ~ 0.015
G is the gelbstoff absorption at 440nm
a    a0    a1  ln PP
a0   and a1   taken from tabulated values in Lee et all.
P0
P1
is the phytoplankton absorption coefficient at 400 nm
which varies with the CHLOROPHYLL concentration.
is dependent on P0
31
Bio-Optical Model 4
 440 
bbp ( )  X 
  
y
Particulate scatter
X is the backscattering coefficient of particulates at 440 nm
y gives an indication of the size particles.
 B  B sd  
Water bottom (lambertian)
Using sand based normalized spectral
response
The parameters in the reflectance model to be retrieved are:
P, G , X , Y , H , B
32