Development and assessment of leaf area index algorithms
for the Sentinel-2 multispectral imager
Richard Fernandes1, Marie Weiss2, Fernando Camacho3, Beatrice Berthelot4, Fred Baret2, Riccardo Duca5
1: CCRS, Government of Canada; 2: INRA, France; 3: EOLAB, Spain; 4: Magelllium, France; 5: ESTEC, European Space Agency
Objectives
 Description of Validation Sentinel 2 (VALSE2)
experiment - focus on LAI algorithm validation.
 Description of two LAI algorithms applicable to
S2 MSI
 INRA Neural Network inversion of PROSAILH
 CCRS Red-Edge analytical solution
Sentinel 2 Mission Requirements
VALSE2 – Review of Algorithms
L=low , P=Partial, F=Full satisfaction or Mission requirements.
Fernandes et al., VALSE2 Algorithm Survey, CCRS, 2014.
INRA NNET Algorithm
Baret et al., VALSE2 CFI Algorithm Theoretical Basis Document, INRA, 2014.
CCRS LAI Algorithm
Continuous radiative transfer equation:
,
,
,
,
,
Interaction
Eigenfunction decomposition:
Scattering
,
,
,
,
,
,
The probability a photon recollides in the canopy
at the infinite scattering order.
Why do we care about p?
,
is invariant to angular or spectral variation of
is analytically related to LAI (Stenberg, 2006)
1
9
0.9
8
Clumping
0.8
0.6
p
7
0.60
0.65
0.70
0.75
0.80
0.85
0.90
0.95
1.0
1.05
0.7
0.5
Erectophile
Planophile
Uniform
6
5
LAI


4
0.4
3
0.3
2
0.2
1
0.1
0
0
2
4
6
8
LAI
Fernandes and Gitelson, submitted to RSE.
10
0
0
1
2
3
4
5
6
7
[1-exp(-clumping*LAI)]*[a +a /(1-p)]
0
1
8
9
Relating p to S2 MSI reflectance
is a function of
(1) black soil reflectance
(2) leaf albedo
Red-edge NDVI for S2 closely related to
=
.
0.07
0.06
0.1
0.05
N(725nm)
N(695nm)
0.08
0.06
0.04
0.03
VZA
0.02
+30º
0º
-30º
0.04
0.01
0.02
0.5
0.55
0.6
0.65
0.7
NDVI S2 Band 4, Band 5
0.75
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
NDVI S2 Band 5, Band 6
Fernandes et al., VALSE2 CCRS Red-Edge ATBD, CCRS, 2014.
0.8
and
CCRS LAI Algorithm
LAI Estimation
Inputs
PUTS
B4,B5,B6
urface
ectance
+
unds on
CHL,
OSPECT
meters, ,
soil refl.
,
.
For each , and
soil reflectance
Model leaf albedo,
from CHL using
PROSPECT5
Estimate CHL
from red-edge
CHL index
Estimate soil
reflectance from
regions with
lowest 5% NDVI
Estimate p and LAI
NO
Is p
valid?
YES
Add LAI(p, ) to
solution
Radiative Transfer Verification
INRA NNET
CCRS Red-Edge
CHL g/cm2
10
20
30
40
50
60
70
80
1:1 line
roducer Validation
INRA NNET
(maize)
=34, RMSE=1.13
CCRS Red-Edge
(maize, soybean)
6
5
L A c tual
4
3
2
1
0
0
1
2
3
4
L Estimated
N=300, MAE=0.45
5
6
idation of Sentinel 2: VALSE2
SEN3EXP
SASI
SEN3EXP
CASI
CASI
EAGLE
AHS
SEN3EXP
EAGLE
HYPER
AHS
CEFLES
AHS
L2 Not
Sen2FLEX
SPARC2004
AHS
yes
CASI
SPARC2003
ROSIS
Yes
No
Sen2FLEX
SPARC2003
HYMAP
No
Yes
AHS
AGRISAR
AHS
No
No L2
CEFLES
AGRISAR
ometry
CAIS
VALSE2 Imagery
Yes
No
Yes
Yes
Yes
available
saics
No
No
No
no
no
yes
yes
yes
mporal
Yes
No
No
No
No
Yes
Yes
Yes
Cloud
y
yes
*
*
*
Yes
Yes
Yes
3
3
2
2
1
1
1
diometry
ectral
**)
ESU
ority in the
ocessing
yes
yes
Yes
yes
Yes
Yes
Yes (Barrax)
No San Rossore
No
No
No
barrax
San Rossore
VALSE2 Ground Reference Data
XP
EX
S
AR
LAI
WC
CHL
VALSE2 LAI Validation V1
A NNET
Saturation in retrieval
due to saturation
of input bands
CCRS Red-Edge
CHL significantly
overestmated (>>60ug/cm2)
VALSE2 LAI Validation V2
A NNET
CCRS Red-Edge
SEN3EXP CASI
8
LAI Red-Edge v2
7
6
5
4
3
2
1
0
0
1
2
3
4
5
Ground Measurement
6
7
N=45 R2=-0.53 RMSE=1.76 B=-0.58 S=1.61
et al., VALSE2 Validation Report, EOLAB, 2014.
8
Conclusions
Better co-ordination and careful processing of
reference datasets so radiometry and in-situ
measurements meet product specifications
Need to perform forest validation (BOREAS, Harz)
Sentinel Level 2P implementing NNET and CCRS
algorithms but users must have patience: MODIS
LAI had ~1 version/2 years.
MSI8
MSI6
MERIS10
MSI5
MERIS9
0.14
MSI4
MERIS8
2 MSI and S3 OLCI Red-Edge
1
0.9
0.12
0.7
0.6
0.08
0.5
0.06
0.4
0.04
0.3
0.2
0.02
0.1
0
600
650
700
750
Wavelength (nm)
800
850
0
900
Bi-directional Reflectance
Relative spectral response
0.8
0.1
How did we estimate leaf CHL?
Why does CCRS Red-Edge
ometimes underestimate LAI?
ttawa cal/val site
Equivalent CHL ug/cm2
LAI
70
5
65
4
60
55
3
50
2
45
40
1
35
30
0

Classement Cartes Graphiques Mobile 2014

Motorsporttermine 2014 - Motorsport in Südtirol