1. Leaf Area Index retrieval by inverting SCOPE model
using hyperspectral data over floodplain meadow
Suvarna Punalekara
Anne Verhoefa
Christiaan van der Tolb
Kevin Whitea
a University of Reading, UK
b ITC, University of Twente,
Netherlands

2. FIELD SITE: Yarnton Mead located towards the Northwest side of
Oxford city, UK
The site is declared as Site of Special Scientific Interest
Highly bio-diverse grassland which is being managed
traditionally for almost 1000 years.
Variability in the biophysical parameters
such as LAI, leaf biochemicals, leaf angles
is crucial to understand the functional
dynamics of these communities.
Coupled canopy radiative transfer models
And SVAT models could be highly
beneficial for ecosystem research
SCOPE model
Source : Google Earth

3. LIDFa is coefficient determining leaf angel geometry
Radiative transfer module of SCOPE
PROSAIL = PROSPECT + SAIL
SAIL model:
Canopy radiative transfer model
(Verhoef W., 1985)
Parameters:
Leaf Area Index (LAI)
Leaf inclination distribution function (LIDFa)
Solar geometry
View geometry
- Leaf reflectance and transmittance (PROSPECT)
Canopy hot spot parameter (q)
Soil reflectance
LIDFa is coefficient determining leaf angel geometry
LIDFa varies from -0.5 to 0.5
-0.5 ~ erect leaves
0.5 ~ planar leaves
q ~ leaf width / canopy height
Output:
Outgoing spectrum of radiation
Hemispherical and in viewing direction
- Net radiation at every surface element

4. PROSPECT: Parameters: Output of Prospect:
Leaf radiative transfer model (Jacquemoud and Baret, 1990)
Parameters:
- Leaf chlorophyll content (μg cm-2) : Lchl
Leaf dry matter content (g cm-2) : Ldmc
Leaf water content (cm) : Lwc
Leaf senescent matter content
(g cm-2) : Lsc
- Leaf thickness parameter (N)
VIS
NIR
SWIR
Output of Prospect:
Leaf reflectance and transmittance
over a range of wavelength

5. Radiative model inversion :
Look-Up Table (LUT) method
General issues:
Which parameters to fix and which one to vary
as free parameters?
What should be the range of parameters?
How to select best fit solution?
LUT dimensions?
Computational time?
Which part of spectrum to be used?
Homogeneous / Heterogeneous target canopies??
Forward model runs
Measured spectra
LUT
Statistical test used for inversion
: Root mean square error (RMSE)
ill-posed
nature of retrieval
Parameters for best fit solution

6. Field Spectro-radiometer
SVC HR-1024i :
- 350 nm nm
- Spectral Resolution: ≤ 3.5 nm,  700 nm, ≤ 9.5 nm,  1500 nm, ≤ 6.5 nm,  2100 nm
- 8o IFOV lens at 1 m height above ground
Retrieval has also been done using airborne hyperspectral data obtained by FENIX

7. Sampling strategy 3 main communities based on dominating species.
10 locations in every community
16 spectra in 1X1m plot at every location
Data processed: FSF toolbox
Carex acuta (CA)
Juncus acutiflorus
(JA)
Sanguisorba
Officinalis (SO) SO JA CA

8. plageophile - near planar
Performance of LUT based retrieval strongly depends on in situ information
(Darvishzadeh et al., 2008 ; Combal et al., 2002)
Field information is highly important in defining parameter space and LUT dimensions
FREE parameters :
LAI,
Leaf chlorophyll content (Lchl)
Leaf dry matter content (Ldmc)
Leaf water content (Lwc)
LIDFa
Hotspot parameter (q)
FIXED parameters:
soil reflectance
Leaf thickness parameter (N)
Leaf senescent matter content
(Lsc)
Sunscan canopy
analyser
Field measurements (13 June 2014)
Lchl
Ldmc
Lwc
LAI q
LIDFa
sps
(µg cm-2)
(g cm-2) cm
m2 m-2
m1 m-1
Unit less SO
32.6 (2.3)
0.006 (0.0003)
0.013 (0.001)
7.88 (1.52)
0.1
plageophile - near planar CA
31.5 (3.9)
0.010 (0.001)
0.018 (0.002)
6.69 (1.03)
0.04
erectophile JA
23.5 (5.1)
0.011 (0.002)
0.033 (0.005)
5.88 (0.94)
0.03

9. Effect of number of combinations selected
RMSE
LAI measured

10. Comparison between modelled ‘best-fit’ and average measured
spectra
All measured wavelengths between 400 and 2470 nm were used
Average of 200 best combinations have been used
Wavelength (nm)

11. Effect of wavelength selection
400:010:2400 nm
400:020:2400 nm
All measured
wavelengths used
400:030:2400 nm
Literature based
LAI meas.
LAI meas.
LAI meas.

12. Effect of fixing one parameter
from the original five free parameters except LAI
Change in RMSE of LAI estimation is compared with the FREE run RMSE (0.78 m2 m-2)
Fixing Lchl and q seem to have improved RMSE
Fixing LIDFa and Ldmc, Lwc seem to make estimations worst.

13. Spectral index based approach for LAI estimation
RMSE for LUT approach : 0.78 m2 m-2
All possible combinations of wavelengths were
Used to calculate NDVI-like index

14. Mask for cloud shadow/target covered areas
FENIX based maps
LAI, (m2 m-2)
Mask for cloud shadow/target covered areas

15. Conclusion
LUT based inversion of PROSAIL model proved to be useful for
LAI estimation for biodiverse floodplain meadow.
Average of 200 ‘best’ solutions retrieved LAI with good RMSE
Reducing number of wavelengths for retrieval did not prove
beneficial for the inversion
Fixing Lchl or q improved the RMSE of LAI retrieval, while fixing
LIDFa, Ldmc and Lwc made it worse.
LUT based approach proved better than spectral index based
approach for LAI retrieval.

16. Thanks !!! Thanks to …… Alasdair Mac Arthur David Macdonald
Ben Marchant
Students and technical staff
from University of Reading
- Anne Verhoef
- Paul Morris
- Irina Tatarenko
- Christiaan van der Tol
Azin Howells