assessment of lai retrieval accuracy by inverting a rt...
TRANSCRIPT
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.