Inland water algorithms

strangerwineAI and Robotics

Oct 19, 2013 (3 years and 11 months ago)

70 views

1

Inland water algorithms

Candidates

and

challenges

for

Diversity

2

Daniel Odermatt, Petra
Philipson
, Ana
Ruescas
, Jasmin Geissler,
Kerstin Stelzer, Carsten Brockmann

2

Chapt

Processing

step

Methods
/
Algorithms
/Tools

Chapter

3:
Atmospheric

Correction

3.6.1

Atmospheric

Correction

over

Land

GlobAlbedo

atmospheric correction

SCAPE
-
M for land

ATCOR 2/3

Durchblick

3.6.2

Atmospheric

Correction

over

Inland

Water

MERIS lakes and C2R

CoastColour

SCAPE
-
M for inland waters

Normalizing at
-
sensor radiances

MERIS lakes and C2R

3.6.3

Adjacency

Effect

Correction

(Water)

ICOL

SIMEC

Context

3

Chapt

Processing

step

Methods
/
Algorithms
/Tools

Chapter

4: Lakes Processing

Pre
-
classification

4.1

Differentiation

of

water

types

Separation by Ecoregion

Optical Water Type classification using Fuzzy Logic

Applicability ranges of water constituents

Quality

4.2.1

IOP retrieval by spectral inversion
algorithms

C2R / CC

FUB

MIP

HYDROPT

Linear matrix inversion approaches

4.2.2

Feature

specific

band

arithmetic

FLH /
MCI

Band ratio
algorithms

4.2.4

Valid

Pixel

Retrieval

Flagging

Statistical filtering

4.2.3

Lakes

Water

Temperature

ARC
-
Lake Data

Set

Quantity

4.3.1

Water

Extent

SAR Water Body Data Set

4.3.2

Lakes

Water

Level

ESA river and lake dataset

Context

4


Introduction


Algorithm requirements


State of the art



Atmospheric correction


Normalizing TOA radiances


Aerosol retrieval over water:

C2R


Aerosol retrieval over land:

Scape
-
M



Water constituent retrieval


Band ratios


C2R



Conclusions

Content

5


Atmospheric correction requirements


Unsupervised, automatic processing


Computationally feasible


Validated use with MERIS images



Water constituent retrieval requirements


… as above


Applicable to optically deep inland waters


Chlorophyll
-
a and suspended matter
products

Introduction

6

Introduction



mg/m
3

g/m
3

m
-
1

1

0
-
3

0
-
3

0
-
0.8

2

3
-
10

3
-
30

0.8
-
2

3

>10

>30

>2

7

Schroeder (2005), Binding et al. (2010), Matthews et al. (2010)


Over clear waters, L
atm

exceeds

L
w

by an order of magnitude



Needs correction



Over very turbid waters, signal

strength increases strongly




Uncorrected analysis possible



Recent studies achieved better

results with FLH, MCI on L
w


Normalizing TOA radiances

8

Doerffer & Schiller (2008)


Explicit atmospheric correction: MERIS Lakes ATBD


TOA is corrected for pressure and O3 variations with MERIS metadata


TOSA consists of aerosols, cirrus clouds, surface roughness variations


BOA RL
w

is calculated by a forward
-
NN


2 different models for atmosphere and water training dataset



C2R background reflectance training range:


0.01
-
100 g/m
3

TSM


0.01
-
43 mg/m
3

CHL


0.003
-
9.2 m
-
1

a
CDOM



Extended CoastColour training range:


1000 g/m
3

TSM


100 mg/m
3

CHL


Aerosol retrieval over water: C2R

9

Odermatt et al. (2010)



Aerosol retrieval over water: C2R

10

Guanter et al. (2010)


Scape
-
M


Modtran

compiled LUTs


DEM input for topographic corrections


Retrieving rural aerosols and water
vapour

over land


Using 5 vegetation
-
soil mixture pixels per 30x30 km


Atmospheric properties are interpolated over lakes


Max. 1600 km
2

lake area and 20 km shore distance

Aerosol retrieval over land: Scape
-
M

11

Guanter

et al. (2010)



Aerosol retrieval over land: Scape
-
M

12


Automatic and accurate correction required in most cases



Aerosol retrieval over water


Provides RT
-
based, accurate estimates for low
reflectances


Application limited by training ranges



Aerosol retrieval over land


Valuable backup for certain niches, e.g. retrieval of secondary
chl
-
a
peak


Limited by atmospheric, geographic and
limnic

constraints

Summary: Atmospheric correction

13


Derived

through

empirical

regression

or

bio
-
optical

modeling



Retrieve

CHL, TSM, CDOM
individually



Primary

CHL
-
absorption

(OC)
algorithms

(400
-
550 nm)
not

applicable



Secondary

CHL
-
absorption

feature

shifting

with

concentration

(681 nm)

Band ratio algorithms

14

Gitelson

et al., 2011:
Azov

Sea



Red
-
NIR

band
ratios

for

CHL

15

Doxaran

et al. (2002):
Gironde

estuary
, Bordeaux



TSM sensitive
bands

a

13 g/m
3

b

23 g/m
3

c

62 g/m
3

d

355 g/m
3

e

651 g/m
3

f

985 g/m
3


16

Watanabe et al., 2011;
Kusmierczyk
-
Michulec

& Marks, 2000



CDOM
absorption

properties

Maritime vs.
inland

aerosols

Maritime vs.
Baltic


sea

aerosols


Ambiguity

can

occur

with

all
other

optical

parameters



Angstrom

variations

over

continents




Band
ratios

make

use

of 2
-
4
bands

of
the

visible

spectrum



Methodological

c
onvergence

is

not

significant

17

Odermatt et al. (2012)

To which optically complex waters do recent “Case 2” algorithms apply?



The literature review includes:


Matchup

validation studies


Constituent retrieval from
satellite

imagery


Optically
deep and complex
waters


Explicit
concentration

ranges and
R
2


Published in
ISI listed
journals


Between Jan
2006

and May
2011

These criteria apply to a total of 52 papers.

Algorithm

validation

ranges

review

18

Odermatt et al. (2012)

The literature review
aims to:


Quantify the use of recent algorithms and sensors


Derive algorithm
applicability ranges

for coastal and inland waters


Clarify the
ambiguous use of attributes
like “
turbid” and “clear


Algorithm

validation

ranges

review

Authors

Oligotrophic

Mesotrophic

Eutrophic

Hypereutr.

Chapra

&

Dobson

(
1981
)

0
-
2.9

2.9
-
5.6

>5.6

n.a.

Wetzel

(
1983
)

0.3
-
4.5

3
-
11

3
-
78

n.a.

Bukata

et

al
.

(
1995
)

0.8
-
2.5

2.5
-
6

6
-
18

>18

Carlson

&

Simpson

(
1996
)

0
-
2.6

2.6
-
20

20
-
56

>56

Nürnberg

(
1996
)

0
-
3.5

3.5
-
9

9
-
25

>25

This

study

0
-
3

3
-
10

>10

?

19

Odermatt et al. (2012)



CHL band
ratios

5
SeaWiFS

2 MODIS

1 GLI

8 MERIS

2 MODIS

1 HICO

2 TM/ETM+

1 MERIS

20

Odermatt et al. (2012)



TSM band
ratios

5
empirical

5 semi
-
analytical

21

Odermatt et al. (2012)



CDOM band
ratios

22

Odermatt et al. (2012)



Spectral

inversion

algorithms

Authors

Algorithm

CHL
[mg/m
3
]

TSM
[g/m
3
]

CDOM
[m
-
1
]

max

min

max

min

max

min

Binding

et

al
.

(
2011
)

NN algal_2

70.5

1.9

19.6

0.8

7.1

0.5

Cui

et

al
.

(
2010
)

NN algal_2

16.1

0.7

67.8

1.5

2.0

0.7

Minghelli
-
Roman

et

al
.

(
2011
)

NN algal_2

9.0

0.0

-

-

-

-

Binding

et

al
.

(
2011
)

NN C2R

70.5

1.9

19.6

0.8

7.1

0.5

Giardino

et

al
.

(
2010
)

NN C2R

74.5

11.7

-

-

4.0

1.3

Matthews

et

al
.

(
2010
)

NN C2R

247.0

69.2

60.7

30.0

7.1

3.4

Odermatt

et

al
.

(
2010
)

NN C2R

9.0

0.0

-

-

-

-

Schroeder

et

al
.

(
2007
)

NN FUB

12.6

0.1

14.3

2.7

2.0

0.8

Shuchman

et

al
.

(
2006
)

coupled

NN

2.5

0.1

2.7

1.3

3.5

0.0

Giardino

et

al
.

(
2007
)

MIM

2.2

1.3

2.1

0.9

-

-

Odermatt

et

al
.

(
2008
)

MIP

4.0

0.6

-

-

-

-

Santini

et

al
.

(
2010
)

2
step

inv

5.0

1.8

13.0

3.0

0.8

0.1

Van

der

Woerd

&

Pasterkamp

(
2008
)

Hydropt

20.0

0.0

30.0

0.0

1.6

0.0

Validation of C2R/algal_2/(FUB):


Numerous and independent


Adequate for low to medium concentrations


Inadequate for high concentrations

Validation of other algorithms:


Limited in number and independence


Often restricted
to “domestic”
use

validated
|
falsified
| threshold R
2
=0.6

23

Odermatt et al. (2012)



Variability

range

scheme

medium

low

high

c
oncentration

level


type


contravariance







Reading
example
:

D‘Sa

et al. (2006)

r
etrieve

low

w
ith

510, 565
nm

bands

at

0.3
-
13.0 mg/m
3

CHL

and

0.5
-
5.5 g/m
3

TSM

TSM

CDOM

510, 565
nm

D‘Sa

et al., 2006

24

Odermatt et al. (2012)



Variability

range

scheme

red
-
NIR band
ratios

for

very

turbid

TSM



red
-
NIR
band
ratios

for

eutrophic

CHL




NN
for

intermediate
concentrations





OC band
ratios

for

oligotrophic

CHL






Representing

coastal

waters

of

mostly

co
-
varying

constituents










band
ratios

for

CDOM

25

wc

retrieval
:

-
FLH, MCI

-
Gitelson

2/3
-
band


atm
.
c
orrection
:

-
none

-
SCAPE
-
M





wc

retrieval
:

-
FUB

-
blue
-
green

bands


atm
.
correction
:

-
C2R

(+ICOL!)

-
FUB

(+ICOL?)





wc

retrieval

&

atm
.
correction
:

-
C2R

-
FUB

Diversity

recommendations

26

Conclusions from the validation review:


Algorithm validity ranges are defined at high confidence (52 papers)


MERIS neural networks are sufficiently and independently validated


MERIS’ 708 nm band provides unparalleled accuracy for eutrophic waters


Open issues for use of the findings in
diversity 2
:


How

is the required
preclassification

applied?


Based

on
previous

knowledge

or

on
-
the
-
flight
?


Spatio
-
temporally

static

or

dynamic
?


based

on
previous

knowledge

or

iterative
processing
?


Should algorithm blending be applied as suggested by
Doerffer

et al. (2012)?

Conclusions

27

Thank

you

28

Schroeder et al. (2007), Schroeder (2005)


Implicit

atmospheric

correction
: FUB


Coupled

water
-
atmosphere

RT
model

MOMO


Simulation of 5
optical

thicknesses

of 8
aerosol

types
, 4 rel.
humidities


Inversion of TOA
reflectance



where

RS
TOA

is

TOA
reflectance

in 12 MERIS
bands

x, y, z
are

transformed

observation

angles

θ
s

is

the

illumination

angle

P
is

surface

pressure

for

Rayleigh
correction

W
is

wind
speed

T
is

transmissivity

And x
is

a
neural

network

learning

pattern

corresponding

to a
set

of
concentrations



(
Originally

not
meant

for

use

for

inland

waters
,
e.g
.
altitude
))



Aerosol retrieval over water

29


CoastColour



Rio de la Plata

IGARSS * Munich * 24.07.2012

CoastColour AC

L1b RGB

SeaDAS l2gen

L2 3
rd

reprocessing

Case2R

L1b band 13 (865nm)

30

Doerffer

et al. (2012)



CoastColour

neural

network


Inv

NN t
rained

with

6

Mio
.
Hydrolight

simulations

Rw(10
bands
)



I佐O



5
IOPs
:
a_pig
,
a_gelbstoff
, b_
particle
; NEW:
a_detritus
,
b_white




SIOP
variations

by

additional
rather

than

hypothetically

accurate

IOPs



Accounting

for

T=0
-
36
°
C, S=0
-
42
ppt



Random

variation

spacing

for

bulk

a and b
instead

of
individual

IOPs



CHL = 21*apig
1.04


for

0.01
-
100 mg/m
3



TSM = (bp1+bp2)*1.73

for

0.01
-
1000 g/m
3



CDOM=
ad+ag


for

0.01
-
4 m
-
1


k
d

calculated

from

IOPs

for

all 10
bands



z
90

is

the

average

of
the

3
lowest

k
d



Corresponds

roughly

to
Secchi

disc

depth


31

Schroeder, 2005; Schroeder et al., 2007


Some

conceptual

differences



Uses

MERIS level1 B TOA

bands

1
-
7, 9
-
10, 12
-
14



Static

3
-
component IOP
model



Forward
simulations

based

on

coupled

atmosphere
-
ocean

model

(MOMO)



NN
for

direct

RS
TOA



IOP
inversion



NN
for

indirect

RS
TOA



RS
BOA



IOP

performed

similarly


FUB
neural

network

32


Alternative architechture


Uses MERIS level1 B TOA bands 1
-
7, 9
-
10, 12
-
14




Schroeder, 2005; Schroeder et al., 2007


Bio
-
optical

model

training

ranges


CHL:

0.05
-
50 mg/m
3


TSM:

0.05
-
50 g/m3


CDOM:

0.005
-
1 m
-
1



Partially

covarying

constituents

assumed







FUB
neural

network

33

Odermatt et al., 2012


Greifensee 2011 EUT/C2R/FUB



16 MERIS
images

within

69
days



CHL
profiles

acquired

automatically



Stratified

cyanobacteria

blooms

Validation
examples