biohype: what can we learn from hyperspectral solar ...€¦ · • chl fluorescence, once emitted...
TRANSCRIPT
BIOHYPE: What can we learn
from hyperspectral solar-induced
chlorophyll fluorescence of urban
tree leaves?
Shari Van Wittenberghe1, Luis Alonso3, Jochem Verrelst3, Inge Hermans2,
Frank Veroustraete1, Roland Valcke2, Jose Moreno3, Roeland Samson1
BEO day, 20th of November 2014, Lier
1Laboratory of Environmental and Urban Ecology, Dep. Bioscience Engineering,
University of Antwerp 2Labboratory of Molecular and Physical Plant Physiology, University of Hasselt
3Image Processing Laboratory, Dep. Earth Physics and Thermodynamics,
University of Valencia
Save BELSPO!
Introduction on chlorophyll fluorescence
Methodology: leaf vs. Remote sensing
Results 1: Drivers of Chl fluorescence
Results 2: Parameter retrieval from
hyperspectral reflectance, absorbance &
fluorescence dataset
Results 3: Chlorophyll content map of urban
vegetation in Valencia
General outcome
2/20
BIOHYPE: What can we learn from hyperspectral solar-
induced chlorophyll fluorescence of urban tree leaves?
Wavelength (nm)
400 600 800 1000 1200 1400 1600 1800 2000 2200 2400
Incom
ing r
adia
nce (
W m
-2 s
r-1 n
m-1)
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
Incoming solar radiance
Wavelength (nm)
400 600 800 1000 1200 1400 1600 1800 2000 2200 2400
Incom
ing r
adia
nce (
W m
-2 s
r-1 n
m-1)
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
Incoming solar radiance
Absorbed leaf radiance
Visible
UV Near Infrared Shortwave Infrared
Energy dissipation mechanisms
(Rosema et al., 1998, RSE)
General outcome Introduction Methodology R1:Drivers of Chl F R2: Parameter retrieval R3: Chl map
3/20
General outcome Introduction Methodology R1:Drivers of Chl F R2: Parameter retrieval
Excitation De-excitation
Chl
S1
S2
1Chl*
heat loss
Fluorescence
S0
Ground
state
Red photon Blue photon
Chl
Wavelength (nm)
400 500 600 700 800 900 1000 1100
Absorb
ed r
adia
nce (
W m
-2 s
r-1 n
m-1
)
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35VISIBLE NEAR-INFRAREDUV
Absorbed light
Wavelength (nm)
400 500 600 700 800 900 1000 1100
Ab
so
rbe
d r
ad
ian
ce
(W
m-2
sr-1
nm
-1)
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
Flu
ore
scence r
adia
nce (
W m
-2 s
r-1
nm
-1)-2
sr-1
nm
-1)
0.000
0.002
0.004
0.006
0.008
0.010VISIBLE NEAR-INFRAREDUV
Absorbed light
Emitted fluorescence
Some vegetation physiology…
R3: Chl map
4/20
Chl F: 650-850 nm
Wavelength (nm)
400 500 600 700 800 900 1000 1100
Appare
nt
reflecta
nce (
-)
0.0
0.1
0.2
0.3
0.4
0.5
Reflectance
+
Chl fluorescence
= Apparent reflectance
• Chl fluorescence is a weak
signal compared to leaf
reflectance
Wavelength (nm)
400 500 600 700 800 900
Ra
dia
nce
(W
m-2
sr-1
nm
-1)
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
Apparent reflected radiance• Cut off filter at 650 nm
Wavelength (nm)
400 500 600 700 800 900
Ra
dia
nce
(W
m-2
sr-1
nm
-1)
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
Apparent reflected radiance
Chl fluorescence
• FluoWat leaf clip (by L.
Alonso) coupled with ASD
spectroradiometer
• Pointed towards the sun
(clear sky is needed)
• Upward and downward Chl
fluorescence emission, together
with reflectance and
transmittance
Spectroradiometer
• Why solar-induced?
General outcome Introduction Methodology R1:Drivers of Chl F R2: Parameter retrieval R3: Chl map
5/20
• Remote vegetation monitoring through
reflectance since „70s
Source: http://ipl.uv.es/flex-parcs/
GOME satellite (Joiner et al. 2013)
• Chl fluorescence: more physiological signal
allowing direct diagnose of actual vegetation status
(instead of “greenness”)
• FLEX (FLuorescence Explorer) mission and
CarbonSat now contesting to be ESA‟s 8th Earth
Explorer, both address key climate and
environmental change issues
(Meroni et al. 2009) Fraunhofer Line Depth
Hyplant sensor – fields with different
fluorescence intensity
General outcome Introduction Methodology R1:Drivers of Chl F R2: Parameter retrieval R3: Chl map
6/20
Fluorescence intensity
Absorbed PAR
• Chl fluorescence intensity ~ solar
intensity
• By pointing towards the sun: constant
sun angle irradiance more constant
• Not interested in Chl fluorescence
intensity, but in Chl fluorescence yield
• Fluorescence Yield (FY) =
• Upward ↑FY, Downward ↓FY
• Total FY= ↑FY + ↓FY
• FY at peak wavelength: 687 nm, 741 nm
• Peak ratios: eg ↑ FY(687)/ ↑ FY(741)
• Bidirectional F ratio: eg ↓FY(687)/↑ FY(687)
Wavelength (nm)
650 700 750 800 850
FY
(-)
0
1e-5
2e-5
3e-5
4e-5
↑ FY
↓FY
General outcome Introduction Methodology R1:Drivers of Chl F R2: Parameter retrieval R3: Chl map
7/20
(Miller et al. 2005, ESA contract report)
Results 1: Drivers of Chl fluorescence
BIOHYPE Dataset
• 4 tree species
• 40 trees, 10 locations
• ≠ sampling heights
• ≠ traffic exposure
• > 300 leaf spectra
General outcome Introduction Methodology R1:Drivers of Chl F R2: Parameter retrieval R3: Chl map
8/20
1. Chlorophyll content
Wavelength (nm)
400 500 600 700 800 900 1000 1100
Ab
so
rbe
d r
ad
ian
ce
(W
m-2
sr-1
nm
-1)
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
Flu
ore
sce
nce
ra
dia
nce
(W
m-2
sr-
1 n
m-1
)-2 s
r-1 n
m-1
)
0.000
0.002
0.004
0.006
0.008
0.010VISIBLE NEAR-INFRAREDUV
Absorbed light
Emitted fluorescence
• Red peak overlaps Chl absorption
range: re-absorption effect
• Red/Far-red peak ratio is sensitive
to Chl: re-absorption in the red, not
in the far-red
• Re-absorption effect stronger for ↓FY
Wavelength (nm)
660 680 700 720 740 760 780
F/
F
0.0
0.2
0.4
0.6
0.8
1.0
re-absorption effect
C. australis
M. alba
P. x acerifolia
P. canariensis
Introduction Methodology R1:Drivers of Chl F R2: Parameter retrieval R3: Chl map General outcome
9/20
0,0
0,5
1,0
1,5
2,0
2,5
3,0
↑ F
Y (
x 1
0-5
)
0,0
0,5
1,0
1,5
2,0
2,5
3,0
↑ F
Y (
x 1
0-5
)
0,0
0,5
1,0
1,5
2,0
650 700 750 800 850
↓ F
Y (
x 1
0-5
)
Wavelength (nm)
0,0
0,5
1,0
1,5
2,0
650 700 750 800 850
↓ F
Y (
x 1
0-5
)
Wavelength (nm)
P. x. acerifolia P. canariensis **
*
**
** **
**
*
2. Air pollution
High traffic exposure Low traffic exposure
• Red/Far-red peak
ratio increases for
high exposure
• No significant
differences in Chl
between classes
• Chl F more
sensitive than
“static” Chl
concentration
Re-absorption ~
Chl efficiency to re-
absorb light
(Van Wittenberghe et al., 2013, Environmental Pollution)
Introduction Methodology R1:Drivers of Chl F R2: Parameter retrieval R3: Chl map General outcome
10/20
3. PS Stoichiometry
P. canariensis
Chl a/b
1.8 2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4
FY
0.0002
0.0004
0.0006
0.0008
0.0010
0.0012
y=0.0006*x - 0.0007 r²=0.8418 (p<0.001)
TS-B
TS-M
TS-T
P. canariensis
Chl a/b
1.8 2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4
FY
(687)/
FY
(741)
0.4
0.5
0.6
0.7
0.8
0.9
1.0
(p<0.01)
TS-B
TS-M
TS-T
A
B
• Chl a/b is indicator of the size of the
LHCII…
• Chl F driver: ↓ Light ↑LHCII ↓ Chl a/b
• LHCII= light harvesting complex associated
with PSII
• LHCII collects solar radiation, and transfers
the excitation energy towards the PSII reaction
centers
• PSII contributes most of the total fluorescence
emission
• No “shadow” leaves, leaf deposit is
suggested as indirect factor
(Van Wittenberghe et al., 2014a, Science of the Total Environment)
Introduction Methodology R1:Drivers of Chl F R2: Parameter retrieval R3: Chl map General outcome
11/20
C. australis M. alba
P. canariensis P. x acerifolia
1
2
3
4
5
5
4. Leaf structure (scattering behavior)
(1) Upper epidermis, (2) Palisade parenchyma, (3) Spongy parenchyma,
(4) Lower epidermis, (5) Mesophyll
equifacial
bifacial bifacial
bifacial
Introduction Methodology R1:Drivers of Chl F R2: Parameter retrieval R3: Chl map General outcome
12/20
Multiple upward
scattering →reflectance
Multiple downward scattering
→ transmittance
4. Leaf structure (scattering behavior)
The same happens with
fluorescence…
Wavelength (nm)
650 700 750 800
Flu
ore
scence Y
ield
(-)
0
1e-5
2e-5
3e-5
4e-5
Flu
ore
scence Y
ield
(-)
0
1e-5
2e-5
3e-5
4e-5
5e-5
Total
Upward
Downward
Wavelength (nm)
650 700 750 800
C. australis M. alba
P. x acerifoliaP. canariensis
• Each species characteristic F
shape and FY
• Same FY ≠ same F shape
• ↓FY up to 40% of FYtot
Introduction Methodology R1:Drivers of Chl F R2: Parameter retrieval R3: Chl map General outcome
13/20
4. Leaf structure (scattering behavior)
equifacial bifacial
• Influence of leaf scattering effects onto the bidirectional partitioning of Chl fluorescence
• Chl fluorescence, once emitted from Chl a, interacts as light of the same wavelength
would do
• However, still variation due to different light origin:
Transmittance ↓F (Van Wittenberghe et al., Accepted, Remote Sensing of the Environment)
Introduction Methodology R1:Drivers of Chl F R2: Parameter retrieval R3: Chl map General outcome
14/20
There is a lot information captured at leaf level which will help to understand Chl F
behaviour at canopy level…
Chl F will not only decrease with
depth into the canopy (↓illumination),
the spectral composition will also
change
Multi absorption and scattering effects
Ground measurements will play an
important role in separating physiological
from canopy structure effects
A correct estimation of canopy
structural parameters (leaf area index,
leaf angle distribution) will be needed
4. Lessons for RS
Introduction Methodology R1:Drivers of Chl F R2: Parameter retrieval R3: Chl map General outcome
15/20 (Van Wittenberghe et al., Accepted, Remote Sensing of the Environment)
P. x acerifolia
Wavelength (nm)
500 1000 1500 2000
Reflecta
nce (
-)
0.0
0.2
0.4
0.6P. canariensis
Wavelength (nm)
500 1000 1500 2000
Reflecta
nce (
-)
0.0
0.2
0.4
0.6
M. alba
Reflecta
nce (
-)
0.0
0.2
0.4
0.6C. australis
Reflecta
nce (
-)
0.0
0.2
0.4
0.6
• Full range (350-2500 nm)
• All data (4species) put together
for leaf parameter retrieval of
• Specific leaf area
• Leaf water content
• Chl content
• Nitrogen content
• Better use of the hyperspectral
data cube
• Gaussian Processes regression:
• Nonlineair, non-parametric
machine learning algorithm
• Confidence interval
• Band ranking
(Van Wittenberghe et al., 2014b, Journal of Photochemistry and Photobiology B :Biology)
• ARTMO toolbox
• ipl.uv.es/artmo/
• MLRA toolbox
Results 2: Parameter retrieval from hyperspectral reflectance, absorbance &
fluorescence dataset
Introduction Methodology R1:Drivers of Chl F R2: Parameter retrieval R3: Chl map General outcome
16/20
Measured SLA (cm2 g
-1)
40 60 80 100 120 140 160 180
Estim
ate
d S
LA
(cm
2 g
-1)
40
60
80
100
120
140
160
180
RMSE=8.5 cm2
g-1
NRMSE=7.2%
1:1
Wavelength (nm)
500 700 900 1100 1300 1500 1700 1900 2100 2300
Fre
qu
en
cy
0
5
10
15
20
Absorb
ance (
-)
0.0
0.4
0.8
710
470
22501490
Fre
qu
en
cy
0
5
10
15
20
Reflecta
nce (
-)
0.0
0.2
0.4
0.6510
910
Specific Leaf Area1310
1490
1730
2170
1310 1730
2190
Rank 1 Rank 2 Rank 3 Rank 4 Rank 5 Rank 6 Rank 7 Rank 8
• Parameter Specific leaf area=fresh leaf area/ dry weight
• Range: 35.1-160.0 cm² g-1
• Spectral input: every 20 nm of full range 86 hyperspectral bands
• 70% training (179), 30% testing (76) data set
• 20 model runs Band selection
Measured SLA (cm2 g
-1)
40 60 80 100 120 140 160 180
Estim
ate
d S
LA
(cm
2 g
-1)
40
60
80
100
120
140
160
180
RMSE=7.3 cm2 g
-1
NRMSE=6.0%
1:1
• Several distinct wavelengths are chosen over the full range
• Related to leaf productional or structural biochemicals
Cellulose,
lignin
Protein
Starch
Protein Pigment
Cellulose,
lignin
Results 2: Parameter retrieval from hyperspectral reflectance, absorbance &
fluorescence dataset
Model performance
(testing dataset)
Introduction Methodology R1:Drivers of Chl F R2: Parameter retrieval R3: Chl map General outcome
17/20
• Parameter: Chl content
• Range: 15.9-189.2 µg cm-²
• Estimation of high Chl content rather difficult
Band selection Model performance
• Red edge most important band
• Absorption maxima (e.g. pigment) avoided: deeper light penetration
Starch
• Other bands related to leaf productional or structural biochemicals
Fre
qu
en
cy
0
5
10
15
20
Re
fle
cta
nce
(-)
0.0
0.2
0.4
0.6
410
510 590 710
950
Chlorophyll a+b
500 700 900 1100 1300 1500 1700 1900 2100 2300
Fre
qu
en
cy
0
5
10
15
20
Ab
so
rba
nce
(-)
0.0
0.4
0.8
1130
710950
590510
2250
1650,1670
(1430)
(1730)
Rank 1 Rank 2 Rank 3 Rank 4 Rank 5 Rank 6 Rank 7 Rank 8
Measured Chl a+b (µg cm-2)
0 50 100 150 200
Estim
ate
d C
hl a+
b (
µg
cm
-2)
0
50
100
150
200
RMSE=14.3 µg cm-2
NRMSE=9.2%
1:1
Measured Chl a+b (µg cm-2)
0 50 100 150 200
Estim
ate
d C
hl a+
b (
µg
cm
-2)
0
50
100
150
200
RMSE=14.4 µg cm-2
NRMSE=9.1%
1:1Pigment Red edge
Cellulose
Cellulose,
lignin
Water
Protein
Starch
Results 2: Parameter retrieval from hyperspectral reflectance, absorbance & fluorescence dataset
Introduction Methodology R1:Drivers of Chl F R2: Parameter retrieval R3: Chl map General outcome
18/20
Using several spectral features, both directly and indirectly linked to a parameter
through covariation, helps avoiding saturation effects
• NAOC (Normalized Area Over reflectance Curve)
• Less saturation compared to common VI:
Results 3: Chlorophyll content map of urban vegetation in Valencia
(Delegido et al., 2014, Ecological Indicators)
• Relation with traffic intensity/habitat quality:
Introduction Methodology R1:Drivers of Chl F R2: Parameter retrieval R3: Chl map General outcome
• CASI, experimental configuration • 144 spectral bands using binning outside areas of interest
• FWHM: 2.4 nm at O2 absorptions and red-edge (630-802 nm) and PRI
bands
• Pixel size Along track: 1.6 m; Across track: 1.0 m
a =643nm, b=795nm
19/20
General outcome of the project
• 7 peer reviewed publications
• Largest/most complete solar-induced leaf
fluorescence dataset published
• Dataset could be further used for modelling
studies
• Results picked up and used by FLEX-team
• BIOHYPE Special Session organized at
• 2 PhDs completed with complete/partial
BIOHYPE support, 1 still running
Introduction Methodology R1:Drivers of Chl F R2: Parameter retrieval R3: Chl map General outcome
Thank you for you attention!
20/20