Assessment of LAI retrieval accuracy by inverting a RT model and a simple a RT model and a simple empirical model with empirical model with multiangularmultiangular and and hyperspectral hyperspectral CHRIS/PROBA data from CHRIS/PROBA data from

SPARCSPARC

3rd ESA CHRIS/Proba Workshop

21 - 23 March 2005

F. Vuolo, L. Dini, G. D’UrsoDept. Agricultural Engineering and AgronomyUniversity of Naples Federico II, Italy

Aim :Aim :

Assessment of retrieval accuracy by using :- RT models and empirical approaches (i.e. veget. indexes)- multi-angular and/or super spectral info

Retrieval of canopy parameters (in particular LAI) fromE.O. data for :- calculation of crop transpiration and soil evaporation(P-M approach)- soil water balance simulations (input forcing)

Present objectivePresent objective ::

Bio-physical models include vegetation parameters

( ) 87.52 /1(1 / )

ns nl E ap

c a

R R G D rEr r

ρλ γ

∆ − − +=∆ + +

(1 )ns tR r S= −

0.5t

cRrLAI

=

2 23 3ln ln

0.123 0.0123

0.168

U c T c

c c

a

z h z hh h

rU

− − =

Based on the definition of crop water requirements of

F.A.O. Paper 56 (1-step approach - Penman-Monteith equation):

Outline of procedures adopted for retrieving LAI

1. SEMI-EMPIRICAL MODEL: CLAIR based on the WDVI (Clevers, 1989)

2. INVERSION OF RT MODEL PROSPECT and SAILH (Jacquemoud & Baret, 1990; Verhoef, Kuusk,1985) WITHOUT A-PRIORIINFORMATION

3. INVERSION OF RT MODEL WITH A-PRIORIINFORMATION FROM SEMI-EMPIRICAL MODEL

4. OPTIMAL BANDS SELECTION ON BOTH STEPS 2 and 3

Case-study: the SPARC campaignBarrax (Spain, July 2003-2004)

1 ln 1 WDVILAI = - ( - )WDVIα ∞

The value of α was estimated by using 9 and 15 field measurements for the 12th

and 14th of July (SPARC2003)The final value of α was taken in correspondence of the minimum error between observed and estimated LAI

b

bb

b bb

bb b

b bbb

bbb b

bbbbbbbb

bb bbb

b

bb

bb bb

bbb

bb

bb

bbbbbb

bbbbb

b

bbbbbbbbbbb

bbbbbb

bb

b

bbb

bbb

bbbbbb

bbbbb

bbbbbb

b

bb b bb

b

bbb

bbb

LAI measurements

CLAIR model cal. 0.00

1.00

2.00

3.00

4.00

5.00

6.00

Garlic

Alfalfa

Onion

Papav

erOnio

nOnio

nAlfa

lfaOnio

nCornAlfa

lfaAlfa

lfaSug

arbee

tCorn Corn Corn

Sugarb

eet

CornPota

toes

LAI (

LIC

OR

LA

I-200

0)

14/07

Soil line slope: 1

WDVI ∞ : 64

α : 0.4

LAI (Clevers) 14/07/2003 y = 0,9565xR2 = 0,868

0,000

1,000

2,000

3,000

4,000

5,000

0,00 1,00 2,00 3,00 4,00 5,00

MEASURED

ESTI

MA

TED

RMSE = 0.46

12/07

Soil line slope: 0.9

WDVI ∞ : 68

α : 0.47

LAI (Clevers) 12/07/2003 y = 1.0295xR2 = 0.8819

0

1

2

3

4

5

6

7

8

0 1 2 3 4 5 6 7

MEASURED

ESTI

MA

TED

RMSE = 0.59

Calibration of LAI (WDVI) empirical relationship in the Barrax site

14/07 - LAI map from CLAIR model (WDVI, Clevers, 1989)

The empirical relationship has been verified by using 40 independent field measurements. The correlation between measured and predicted LAI, evaluated in terms of root mean square error

0,000

1,000

2,000

3,000

4,000

5,000

6,000

7,000

0,000 1,000 2,000 3,000 4,000 5,000 6,000 7,000

LICOR LAI-2000 - Ground measurements

LAI (

CLA

IR m

odel

)

AlfalfaCornOnionGarlicPotatoSugarBeet1:1

RMSE=0.59

y=0,89xR^2=0,80

Set-up of the RT model inversionIn order to calibrate the model inversion the BRDF of 39 plots were extracted from the CHRIS/PROBA images of the 12th of July by taking the mean reflectance value at each angle in a window of 3x3 pixels

PROSPECT

• The leaf structure parameter N, the leaf chlorophyll a + b concentrationCab, the equivalent water thickness Cw and the dry matter content Cm were bounded to the existence range of the model parameters

SAILH

• The mean leaf inclination angle (LIDF) and the hot spot size-parameter(HOT) were bounded to the existence range of the model parameters

• A standard background reflectance value was used.

Robustness of the inversion procedure of PROSPECT SAILH models applied to synthetic data

0.00

1.00

2.00

3.00

4.00

5.00

6.00

7.00

8.00

9.00

0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00

MEASURED LAI

ESTI

MA

TED

LA

I

AlfalfaPotatoesCornSugarbeetOnion1:1

Potatoes

Alfalfa

Identification of parameters ranges for the inversion procedure (calibration step I) on image of July 12th, 2003

Using a prior information based on initial parameter set((boundedbounded in a in a rangerange close toclose to groundground measurementsmeasurements and and referencereference valuesvalues))

0

1

2

3

4

5

6

7

8

0 1 2 3 4 5 6 7 8

MEASURED LAI

ESTI

MA

TED

LA

I

-15%

+15%

RMSE = 0.42

Identification of parameters ranges for the inversion procedure (calibration step II)

y = 0,6895xR2 = 0,2568RMSE= 1,1

0,00

0,50

1,00

1,50

2,00

2,50

3,00

3,50

4,00

4,50

5,00

0,000 0,500 1,000 1,500 2,000 2,500 3,000 3,500 4,000 4,500 5,000

MEASURED LAI

ESTI

MA

TED

LA

I

NO PRIOR INFORMATION ON CROP TYPE

Validation on image of July 14th, (step I)

PRIOR INFORMATION ON CROP TYPE (initial parameter set)

y = 0,7833xR2 = 0,8374

RMSE= 0,96

0.00

1.00

2.00

3.00

4.00

5.00

6.00

7.00

0.000 1.000 2.000 3.000 4.000 5.000 6.000 7.000

MEASURED LAI

ESTI

MA

TED

LA

I

RMSE = 0.96

CalibrationRMSE=0.42

Validation on image of July 14th, (step II)

0,00

1,00

2,00

3,00

4,00

5,00

6,00

7,00

0,000 1,000 2,000 3,000 4,000 5,000 6,000 7,000

MEASURED LAI

ESTI

MA

TED

LA

I

Alfalfa

Corn

Potatoes

Sugarbeet

Onion

1:1

In order to evaluate the use of prior information and the initial parameter set in the estimation process:

LAI: 1.32

62 spectral bands and 5 view angles

A set of 14 out of 62 (CHRIS mode 1) was selected for each view direction using information based on data quality and references

9 13 26 29 36 40 45 51 56 611511 3925

0

10

20

30

40

50

60

70

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61CHRIS/PROBA BANDS

REF

LEC

TAN

CE

(%)

Optimal band selected

PROSPECTSAILH

CHRIS/PROBA 14/07

in visible (516, 536, 558, 579, 685 nm)in the red edge (702, 747 nm)in the near-infrared (776, 814, 883 nm)

In order to evaluate the the optimal spectral sampling:

In order to evaluate more accurately the optimal spectral sampling, the sensitivity of each observation involved in the parameter estimation process was computed.

2 16 28 32 36 40 46 52383430 4118

2414 25

12

0

10

20

30

40

50

60

70

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61CHRIS/PROBA BANDS

REF

LEC

TAN

CE

(%)

Optimal band selected

PROSPECTSAILH

CHRIS/PROBA 14/07

in visible (447, 548, 568, 587, 611, 669 and 680 nm )in the red edge (697, 709, 721, 734, 748 and 762 nm )in the near-infrared (776, 783, 839 and 893 nm )

Parameter Estimated Lower limit Upper limit Estimated Lower limit Upper limit

N 1,06 0,51 1,61 1,00 0,58 1,48Ca+b (mgcm-2) 40,80 30,80 50,87 31,03 25,95 36,10

Cw (gcm-2) 0,022 0,022Cm (gcm-2) 0,004 0,001 0,006 0,003 0,001 0,005

LAI (m2m-2) 1,21 1,08 1,35 1,30 1,17 1,43HOT 0,001 0 0,001 0,001 0 0,006

Esky (%) 0,16 0,16LIDF 45° 45°

Objective function value

Correlation Coeff.

Iteration number

Total model calls

Time (sec.) 35,47

94,89

0,9981

26

375

60,72

87,39

0,9982

18

262

fixed

fixedfixed

fixed

14 bands from references 17 bands from sensitivity analysis

SPARC 2003

SPARC 2004

Alfalfa canopies from:

LAI: 1.32

LAI: 4.00

Effect of perturbation on a-prior information (initial LAI ±20%)

0,0

0,5

1,0

1,5

2,0

2,5

3,0

3,5

4,0

4,5

5,0

0,0 1,0 2,0 3,0 4,0 5,0

LAI (Measured)

LAI (

Estim

ated

)

14/07-SPARC200316/07-SPARC20041:1

y = 1,01x R^2 = 0,80

RMSE=0,27

0,0

0,5

1,0

1,5

2,0

2,5

3,0

3,5

4,0

4,5

5,0

0,0 1,0 2,0 3,0 4,0 5,0LAI (Measured)

LAI (

Estim

ated

)

14/07-SPARC200316/07-SPARC20041:1

y = 1,01x R^2 = 0,72

RMSE=0,36

0,0

0,5

1,0

1,5

2,0

2,5

3,0

3,5

4,0

4,5

5,0

0,0 1,0 2,0 3,0 4,0 5,0LAI (Measured)

LAI (

Estim

ated

)

14/07-SPARC200316/07-SPARC20041:1

y = 1,02x R^2 = 0,80

RMSE=0,28

Initial LAI(WDVI) –20%

Initial LAI(WDVI) +20%

NO calibration required!

14/07/2003

They were co-registered and all bands stacked in one image. Degraded to a spatial resolution of 50 m to reduce the spatial error among the five images and the inversion time

Classified in 4 different land use in order to add a priori information based on the distribution of some canopy characteristics that can better constrain the inversion process (initial value and bounds)

LAI map derived inverting the PROSPECT SAILH modelsby using CHRIS data

Conclusions

• The data presented here provide good evidence that multi-angular and hyper-spectral remote sensing approaches may have real potential forestimating LAI and other biophysical parameters of agriculture crops.

• In spite of the limitations of the model assumptions the PROSPECT SAILH produced satisfactory results for a wide range of crops.

• The incorporation of a priori information based on estimation of LAI fromWDVI improves inversion results and compensates for initial assumptions of PROSPECT and SAILH models.

• It has been confirmed as in previous studies that the reduction of spectral redundancy inprove results of inversion. Therefore the CHRIS/PROBA mode 3 configuration may appear optimal in heterogeneous landscapes.