Biogeochemistry of the Scheldt estuary and plume

(1)

Airborne

Hyperspectral

Potential for Coastal

Biogeochemistry

of

the

Scheldt

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

(2)

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

(3)

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).

(4)

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

(5)

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 species

Biochemical 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

(6)

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

(7)

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

(8)

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)

(9)

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

%

(10)

0.67

h

Chl a

?

Difficulties of classical algorithms for

Case 2 waters

(11)

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)

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

12

λ:

highest R²

λa/λb

All combinations R² 12 λa/λb: highest R² 42 R² Parameter P1

=A

• +B

• +C

highest R²

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)

(13)

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

.6

7 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

(14)

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









(15)

16

188 µM

R2 _{= 76.6%}

Dissolved Organic Carbon (DOC)



 

₀

0 _.

.

₅₆₆

510 

566 .

0 .

577

0 _C

B

A

DOC









(16)

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









(17)



 

₀

0 _.

.

₄₇₆

465 

476 .

0 .

543

0

2

A

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 %}

(18)

7.25 8.43 R2 _{= 75,00 %}

































487 .

0

841 .

0

487 .

0

589 .

0 C

B

A

pH

(19)

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

(20)

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









(21)

(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)

(22)

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)

(23)

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?

(24)

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)

(25)

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)

(26)

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.