Airborne
Airborne
Hyperspectral
Hyperspectral
Potential for Coastal
Potential for Coastal
Biogeochemistry
Biogeochemistry
of
of
the
the
Scheldt
Scheldt
Estuary and Plume
Estuary and Plume
M. Shimoni & M. Acheroy
Signal and Image Centre, Royal Military Academy;
D. Sirjacobs & S. Djenidi
GeoHydrodynamic and Environmental Research department, Liege University;
A. V. Borges & M. Frankignoulle
Chemical Oceanography Unit, Liege University;
L. Chou
Chemical Oceanography and Water Geochemistry department, Free University
Brussels;
W. Yvermans
Location, Objective and Interest
Airborne campaign
In-situ campaign
Image preparation
Atmospheric correction
Multiple regression
Improvements by across track filtering
Summary
Correlation with first derivative spectrum
Objective:
Explore the possibility of using CASI-SWIR airborne hyperspectroscopy to
retrieve parameters of interest for biogeochemical studies of the Scheldt
Estuary and Plume (i.e. phytoplankton pigments, particulate organic matter,
colour dissolved organic matter and mineral suspended matter) for case-II
ecosystems studies .
Interest:
Develop an approach that may provide high resolution synoptic view of
particular parameters to complete the in situ metabolic data sets (ground
truth obtained from cruises).
CASI-2
SWIR
Field of view (FOV) 37.8 DEG 40 DEG IFOV 1.3 mRAD 1.1 mRAD Spatial resolution 4 m 4 m Spectral range 431-969 nm 850-2500 nm Spectral channel 48 180
The
ai
rbo
rn
e
ca
mpa
ig
n
5 CASI Flight lines
13 In Situ stations
SEPTEMBER 2002
Airborne sensors
Problems: Spectral bands not set as required SWIR affected by calibration problems
Spectrometry (ASD)
1) Field spectrometry records of water leaving radiance;
2) Laboratory reflection and absorption spectrum measurements on collected water samples and on some “isolated” optically active components:
*
total suspended matter,*
colour dissolved organic matter (CDOM),*
identification of main phytoplankton speciesBiochemical parameters
Salinity chlorophyll a Dissolved inorganic carbon (DIC) particulate organic carbon (POC) Temperature phaeopigment Dissolved organic carbon (DOC) particulate inorganic carbon (PIC) pH Total alkalinity Mineral suspended matter particulate nitrogen (PN)
pCO2 Dissolved O2 nutrients (NH4, NO3/NO2, PO4, Si) Pigments by HPLC coloured humic substance (CDOM)
Identification of phytoplankton species
Image Processing
Basic Processing
Radiometrical, Geometrical and Atmospherical
corrections.
Gain and offset corrections using « Effort »
Filtering
Cross track correction
Other processing
Extraction of spectra corresponding to ground truth
stations
Research of correlations between spectral and in-situ
data
Algorithms
Image processing
0 0,05 0,1 0,15 0,2 0,25 0,3 0,35 0,4 0,45 0,5 0 , 4 3 1 0 , 4 4 20 , 4 5 30 , 4 6 5 0 , 4 7 60 , 4 8 7 0 , 4 9 80 , 5 1 0 , 5 2 1 0 , 5 3 20 , 5 4 3 0 , 5 5 50 , 5 6 60 , 5 7 70 , 5 8 9 0 , 6 0 , 6 110 , 6 2 3 0 , 6 3 40 , 6 4 60 , 6 5 70 , 6 6 9 0 , 6 8 0 , 6 9 1 0 , 7 0 3 0 , 7 14 0 , 7 2 60 , 7 3 70 , 7 4 9 0 , 7 60 , 7 7 20 , 7 8 30 , 7 9 5 0 , 8 0 7 0 , 8 18 0 , 8 3 0 , 8 4 1 0 , 8 5 30 , 8 6 40 , 8 7 60 , 8 8 7 0 , 8 9 9 0 , 9 110 , 9 2 2 0 , 9 3 40 , 9 4 50 , 9 5 70 , 9 6 9 wave le ngth c o rr e la ti o n ATCOR 0 0,05 0,1 0,15 0,2 0,25 0,3 0,35 0,4 0,45 0,5 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 2 0 2 1 2 2 2 3 2 4 2 5 2 6 2 7 2 8 2 9 3 0 3 1 3 2 3 3 3 4 3 5 3 6 3 7 3 8 3 9 4 0 4 1 4 2 4 3 4 4 4 5 4 6 4 7 4 8 wave le ngth c o rr e la ti o n FLAASH 0 0,05 0,1 0,15 0,2 0,25 0,3 0,35 0,4 0,45 0,5 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 2 0 2 1 2 2 2 3 2 4 2 5 2 6 2 7 2 8 2 9 3 0 3 1 3 2 3 3 3 4 3 5 3 6 3 7 3 8 3 9 4 0 4 1 4 2 4 3 4 4 4 5 4 6 4 7 4 8 wave le ngth c o rr e la ti o n ACORN
Correlation between in-situ and airborne reflectance
following different atmospheric correction
Time Position In (m) Position Out (m) St.
In Out E° N° E° N°
Distance (m) Velocity (m/min) 4 12h23 12h32 528900 5694740 528136 5694572 782 87 5 12h43 12h55 532116 5695004 531128 5694756 1018 85 8 13h08 13h16 535552 5695562 535084 5695524 469 59 10 13h31 13h42 538880 5697672 538260 5697516 639 58 11 13h51 14h03 537760 5699064 537508 5698840 373 28 12 14h13 14h23 535048 5702960 535036 5702492 470 47 13 14h36 14h46 531408 5706148 531892 5705532 783 78 14 15h01 15h09 528504 5711688 528504 5711196 492 61 9 15h36 15h48 538036 5696616 538192 5696356 303 25 7 15h55 16h06 534796 5697212 535208 5697212 232 21 6 16h14 16h26 531524 5696848 531792 5696724 295 21 3 16h34 16h43 528016 5695436 528248 5695316 261 29
Local averaging considering boat
drifting (initial and final GPS)
Average spectra of each in situ
station from CASI sensor
0
20
40
60
80
100
0,43
0,53
0,63
0,73
0,83
Wavelength (micrometers)
R
e
fl
e
c
ta
n
c
e
%
0.67
h
Chl a
?
Difficulties of classical algorithms for
Case 2 waters
20,00
30,00
40,00
50,00
60,00
0,65
0,67
0,69
0,71
0,73
0,75
Wavelength (micrometers)
R
e
fl
e
c
ta
n
c
e
%
Absence of absorption pic around 0,67 micrometers
12 stations
3
8
W
a
te
r
q
.
p
a
ra
m
ete
rs
4
2
s
p
ec
tr
a
l
b
a
n
d
s
12λ:
highest R²λa/λb
All combinations R² 12 λa/λb: highest R² 42 R² Parameter P1=A
•
+B
•
+C
highest R²
Multiple regressions
Multiple regressions
between
P1
and
all combinations of
P1
ratio 1 (any selected ratios)
λa/λb
ratio 2 (ratio between
any spectral band
and the common denominator)
λc
/λb
(Hirtle & Arencz, IJRS, 24-5, 953-967)
Parameter
Correlation
coefficient (%)
1
Cryptophytes
96.91
2
Dinoflagellates
96.61
3
Diatoms
(*10+6 ind.)72.70
4
PCo
2(ppm)
72.70
5
DIC (mmol/kg)
71.70
6
DOC (µmol)
76
.67
CDOM
(absorb 380 nm)70.22
8
Chl a
48.21
9
Chl c2
42.62
10
Chl b
44.20
Multiple regression results
image range
in situ range
0 . 2 5 t o 2 * 0 . 5 t o 7 . 7
14 - 9 3 4
3 9 0 - 6 8 9
1. 4 5 - 2 . 8 3
2 .0 9 - 2 . 3 9
16 - 18 8
114 - 17 6
1. 6 7 - 11. 3 8
5 - 6 . 9
1.67
11.38
Absobance 380 nm m-1
R2 = 70.3%
Colour Dissolved Organic Matter (CDOM)
0
0
.
.
510
521
521
.
0
.
887
0
C
B
A
CDOM
16
188 µM
R2 = 76.6%
Dissolved Organic Carbon (DOC)
0
0
.
.
566
510
566
.
0
.
577
0
C
B
A
DOC
1.45
2.83 Mmol/kg
R2 = 71.7%
Dissolved Inorganic Carbon (DIC)
0
0
.
.
657
498
657
.
0
0
.
555
C
B
A
DIC
0
0
.
.
476
465
476
.
0
.
543
0
2A
B
C
PCO
DIC
PCO
2
00..657498
657 . 00.555 C B A DIC 1.45 2.83 Mmol/kg R2 = 71,70 %Salinity
476 . 0 465 . 0 476 . 0 543 . 0 C B A Salinity 14 934 ppm R2 = 72,70 % 26 43 R2 = 72,40 %7.25 8.43 R2 = 75,00 %
487
.
0
841
.
0
487
.
0
589
.
0
C
B
A
pH
pH
Across track Filtering
Principle (for each flight line):
Fit a 2nd degree polynomial on the average
accross track signal of each spectral band
Apply a constant filtering to remove this low
frequency signal
Correlation improvements:
before filtering
after filtering before filtering after filtering
DIC
71.7
87.7
1.45 - 2.83
1.89 - 3.90
DOC
76.6
34.9
16 - 188
Gilvin
70.3
77.7
1.67 - 11.38
3 - 14.43
1.89 3.90
Mmol/kg
R2 = 87,8%Dissolved Inorganic
Carbon (DIC)
0
0
.
.
589
876
589
.
0
0
.
841
C
B
A
DIC
(Lahet et. al, IJRS, 2001, 22-9, 1639-1664)
Correlations with First Derivative Spectrums
Principle :
For each of the 38 parameters, run an automatic search for
best correlation with the first derivative
(Rλ
(i+1)
- Rλ
(i)
)/(λ
(i+1)
- λ
(i)
)
Correlation improvements:
Salinity, Si, No3, Po4, Talk, pCO2, DIC:
R² ranging from 75% to 79,5 %, i.e. 3 to 6% higher then previously
All obtained with λ
(i)= 464 nm ; λ
(i+1)= 475 nm
CDOM: R² of 72,5 % (2,5% improvement), with λ
(i)= 442 nm ; λ
(i+1)= 453 nm
SPM:
R² of 41,4%
(or R= 64 % comparable with 60,7 % from Lahet, 2001, same approach, but
higher number of stations, differenciation of water types, lower charge)
Correlations with First Derivative Spectrums
SPM:
ULg versus ULB SPM mg/L
y = 1,0022x + 14,552
R
2= 0,7143
0,00
10,00
20,00
30,00
40,00
50,00
60,00
70,00
80,00
0,00
20,00
40,00
60,00
ULB SPM mg/l
U
L
g
S
P
M
m
g
/l
Ground « Truth » is also affected by errors (systematic and accidental, like for
stations 4 and 14)
Back to Chlorophyll, Multiple Regressions
R² previously obtained: 48,21%
Ratios:
R634/R487 nm
R498/R487 nm
Without problematic samples 4 and
14: improvement to about 58%
Chla model versus measures
y = 0,4348x + 1,8475 R2 = 0,5788 2,00 2,50 3,00 3,50 4,00 2,00 2,50 3,00 3,50 4,00 4,50
Chla in situ (microg/l)
C h la m o d ( m ic ro g /l )
Station 13 out of the line…a confirmation of the necessity to split study
according to water types (improvement to 87%) , or to non linear model?
Summary
Summary
CASI products
SWIR data were suffering important radiometric correction problems and could not be used CASI Spectral bands could not be set as designed
CASI data recorded over water seems affected by
a) “sun glim” and cross track systematic artefacts as revealed clearly by some parameter
distribution (i.e. DIC)
b) Bathymetry effects (i.e. DOC)
First attempt to remove artefacts were realised with cross-track filtering.
Improvement of across track correction method (various parameters to be adapted, artefact of geometrical or radiometrical correction?)
Localisation of zones of extremes values (out of range values) for each parameter, comparison and masking (such as in first flight line, consequently to intense direct sun reflection on water surface, and to the presence of very shallow sand banks probably responsible for extreme values in some ranges)
Summary
Summary
Algorithms and prediction of biogeochemical parameters
Ratio Multiple Regression and well as First Derivative approach proved very convenient for first statistical exploration of large hyperspectral databases
Some results obtained are very encouraging in terms of R², range, distribution (i.e. CDOM) In case of limited number of ground truth stations, attention must be paid to the general
distribution range and pattern established in order to avoid inconsistent correlations
Developing specific algorithms by splitting database according to general water types
Combining previous empirical approach with (a) physical models, including fine bathymetry effect and (b) classical hyperspectral algorithms making use of known isolated spectral signatures (from literature and laboratory spectral measurements on water samples)
Summary
Summary
Constraints of the marine environment
At sea, “Ground truth” may be difficult to define, and requires much time and resources.
Time delay between in situ and flight should be reduced (using several smaller boats), as it is a very dynamic scene.
Ground truth datadbase must be more important in order to allow some validation and accuracy assessment. Enlarging database should be done through (a) increased number of in situ stations; (b) data exchange with other campaigns realised in the same period (TNO netherlands CASI campaign; Mumm, V. de Cauwer).
Hydrodynamic simulation of the water displacement during field campaign may allow to reduce artefacts due to non synoptic measures, and provide deeper understanding of dilution factors and patterns.