Barrax validation agricultural site: Leassons learnt during SPARC campaigns

VALERI meeting, Avignon, 10 March 2005

Barrax validation agricultural site: Leassons learnt during SPARC campaigns

B. Martínez, F. Camacho-de Coca, F.J. García-Haro, A. Verger

UIT- Universitat de Valencia

F. Baret, M. Weiss

INRA - Avignon

VALERI meeting. Avignon, 10 March 2005Universitat

de València


OBJECTIVE

TEST SITE AND FIELD CAMPAIGN

TEST SITE CHARACTERIZATION

- In-situ measurements

- Sampling strategies

TRANSFER FUNCTION: Spatial extension of in-situ measurements

A CASE STUDY WITH LOW RESOLUTION PRODUCTS

CONTENTS


Appropriate Ground Data Set

LAI/FVC/FAPAR

SPARC 2003, 2004

To derive accurate high-resolution maps from in-situ measurements for the validation of (SEVIRI) coarse satellite vegetation products

OBJECTIVE

Medium and Coarse

Biophysical products

Agregation of high -resolution maps

VALERI methodology

Up

-sca

lin

g

High – Resolution biophysical maps

Spatial extension of

local measurements


CONTENTS

OBJECTIVE





TRANSFER FUNCTION: Spatial extension in-situ measurements



Test Site: Barrax (Albacete)

Direct Validation: The Barrax agricultural site of 55 km2 (‘Las Tiesas’) selected for ground measurements acquisitions. All facilities are available.

Indirect Validation: A larger area of 5050 km2 is selected for inter-comparison and validation of SEVIRI products. Very flat area. Crops and natural vegetation.

Two different areas selected:

Natural Vegetation

CropsSoil


The SPARC campaigns are a combination of different initiatives (ESA, CNES, EU, EUMETSAT) but with the common interest of in-situ characterisation simultaneously to airborne and multi-sensors data acquisitions mainly focused on algorithm and product validation.

SPARC’03 from 13th to 14th of July 2003SPARC’04 from 13th to 17th of July 2004

Our participation was funded by LSA SAF (EUMETSAT) !

Field Campaign: SPARC experiments

In situ measurements Available Imagery

Gap Fraction (LAI,FVC, FAPAR)

Chlorophyll

Radiometry

Temperature

Emisivity

Atmospheric profiles

ROSIS (1m) ; HyMAP (5m); AHS (2.5m)

SPOT/HRV (20m)

Landsat/TM (30m)

CHRIS/PROBA (34m)

MERIS/Envisat (300m-1km)

SEVIRI/Meteosat-8 (3 km)


OBJECTIVE






• A CASE STUDY WITH LOW RESOLUTION PRODUCTS


TEST SITE CHARACTERIZATION: Sampling Strategy

Hemispherical camera

Sampling design VALERI methodology.

12 Photographs per ESU

GPS was recorded at the center of the ESU

LICOR LAI2000

Average of 3 replications

24 measurements per ESU

The replications were distributed randomly within the ESU

CROPS FIELDS ESUs_LAI2000 FIELDS ESUs_CAMERA CORN 7 21 4 10

ALFALFA 6 19 3 16 SUGAR BEET 4 19 1 6

ONION 2 16 2 2 GARLIC 1 11 1 2 POTATO 1 14 1 6

VINEYARD 0 0 1 1 GRASS 0 0 1 6

FRUIT TREES 0 0 2 2 TOTAL 21 100 16 53

CROP FIELDS_LAI2000 ESUs_LAI2000 FIELDS_HP ESUs_HP Corn 6 19 6 7

Alfalfa 6 22 6 6 Sugarbeet 2 20 4 5

Onions 3 12 6 8 Garlic 2 17 1 4 Potato 3 16 3 4

Vineyard 1 4 2 4 Sunflower 3 27 1 3

Forest 0 0 1 4 Grass 0 0 1 1 TOTAL 26 137 31 46

Processed with


TEST SITE CHARACTERIZATION: In-situ measurements


1. Retrieved biophysical parameters

TEST SITE CHARACTERIZATION: Results

0

5

10

15

20

25

0 0,8 1,6 2,4 3,2 4 4,8 5,6 6,4

CO

UN

TS

LAI

LAI from LICORSPARC03

0

2

4

6

8

10

0 0,8 1,6 2,4 3,2 4 4,8 5,6

CO

UN

TS

LAI

LAI from HPSPARC03

0

2

4

6

8

10

0 0,8 1,6 2,4 3,2 4 4,8 5,6

CO

UN

TS

LAI

LAI from HP SPARC04

0

5

10

15

20

25

0 0,8 1,6 2,4 3,2 4 4,8 5,6

Cou

nt

LAI

CO

UN

TS

LAI from LICORSPARC04

SPARC’03 SPARC’04

LIC

OR

DH

P



2. Comparison of LICOR and DHP LAI estimates Relative Error between LICOR and DHP mean values per fields.

LAI DHP-LICOR

Relative error typically <25%

FVC DHP-LICOR

Relative error <15% for all cases

-100

-75

-50

-25

0

25

50

75

100

A9 A1 A10 P1 C2 C1 G1 B3 ON1

Err

or

Re

lati

vo

(%

)

CULTIVOS

LAI

SPARC 2003

A1, A10, P1, G1, ON1 and B3 (downward looking)

A9, P1, C2, C1 (upward looking)Camera position:

Large relative error found for DOWNWARD LOOKING PHOTOS !


3. Dependent on the camera position


Er(LAI_DW)= 37% Er(LAI_DW)= 57%

The estimated LAI can be twice in downward looking position !!!

During the SPARC’04 field campaing, all the photographs were taken UPWARD LOOKING when it was possible. Special attention was paid in comparing the results when the camera was in downward and upward looking for some crops.

Sunflower Corn



Attention was paid during SPARC04 in measuring simultaneously with LICOR and DHP. In addition, an intercomparison of different LICORs was done

-100

-75

-50

-25

0

25

50

75

100

SB1 SB1 SB C C ON ON

Er(effective-LICOR)

Er(

%)

CROP

FIELD Er(%)

CORN (C1) 24%

GARLIC (G1) 48%

POTATO (P) 18%

SUGARBEET (SB) 48%

Good agreement between DHP (upward looking) and LICOR estimates.

The largest discrepancies found for dens cover (Sugar Beet) are similar to that shown by different LICOR instruments (around 50%).

HP vs LICORLICOR vs LICOR


OBJECTIVE








TRANSFER FUNCTION: Spatial Extension to High resolution

Transfer Function: Weighted Multiple linear regressions were computed with all possible SPOT bands combinations

The band combination was selected based on the lowest RMSEW (Weighted Root Mean Square Error), RCROSS (Cross Validation RMSE) and weights null, following the methodology proposed by Weiss,(2004)

Instrument (G, R, NIR) INTERCEPT RMSE RMSEW RCROSS

LI-COR -0.011 -0.014 0.021 0.7 0.61 0.54 0.61

HP -0.004 -0.019 0.016 1.5 0.71 0.48 0.70

LAI

Instrument NDVI INTERCEPT RMSE RMSEW RCROSS LI-COR 1.46 0.034 0.13 0.11 0.14

HP 1.41 0.030 0.18 0.17 0.16

FVC

Instrument NDVI INTERCEPT RMSE RMSEW RCROSS HP 1.46 0.1 0.17 0.10 0.18

FAPAR


TRANSFER FUNCTION: Spatial Extension to High resolution

HP LICOR

MEAN 1.91 1.71

STD 1.48 1.62

Correlate 0.99

Bias 0.204

RMS 0.263

FVC

MEAN 0.38STD 0.28

FVC FAPAR

FAPAR

MEAN 0.46STD 0.33

LAI

LICOR HP


Influence of Sampling Spatial Strategy

CONVEX HULL for DHP

Different Samplings Designs

WhiteInterpolatedBlack ExtrapolatedBlue Considering a relative error of the 5%

Transfer Function FLAG IMAGEComputation of convex hull over the collocated radiance values with the in situ measurements

Smallest convex region that contains the data set


OBJECTIVE






A CASE STUDY


Barrax 5x5 km2

0

0,1

0,2

0,3

0,4

0,5

0,6

IN-SITU VGT_LandSAF VGT_CYCL

FV

C mean

std

Barrax test site (5x5 km)

A CASE STUDY: Comparison with large scale products (1 km res)

Barrax 5x5 km2

0

0,1

0,2

0,3

0,4

0,5

IN-SITU MODIS VGT_CYCL

FA

PA

Rmean

std

PRODUCTS (10-days composition):

VGT_LandSAF_v1 SEVIRI Land SAF algorithm (UV) on VGT k0 data

VGT_CYCLOPES_v1

MODIS/TERRA

Barrax 5x5 km2

0

0,2

0,4

0,6

0,8

1

1,2

1,4

1,6

1,8

2

IN-SITU VGT_LandSAF VGT_CYCL MODIS

LA

I

mean

std


Leaf Area Index (LAI) In-situ_3km VGT_CYCLOPES_V1 MODISVGT_LandSAF_V1

Good agreement (RMS<0.25, r2>0.85) for both algorithms LSA SAF_V1 and CYCLOPES_V1 on VGT data

A CASE STUDY: Comparison with large scale productsBarrax large area (50x50 km2)


Fraction of Vegetation Cover (FVC)

In-situ_3km VGT_CYCLOPES_V1VGT_LandSAF_V1

A CASE STUDY: Comparison with large scale products

Good agreement (RMS<0.09,r2>0.7) for the LSA SAF_v1 algorithm on VGT data

Barrax large area (50x50 km2)


In-situ_3km MODISVGT_CYCLOPES_v1

Fraction of Absorbed PAR (FAPAR)

A CASE STUDY: Comparison with large scale products

Good agreement (RMS<0.08, r2>0.8) for MODIS product is found


CONCLUSIONS

Large amount of ground and airborne data has been collected in Barrax from 1998 (DAISEX) up to now (SPARC, DEMETER), and two new ESA field campaigns are planned for 2005 (June and July).

DHP is used for LAI, FVC and FAPAR estimates in addition to LICOR. The pre-processing with CANEYE has been found quite independent of the operator (UV vs INRA).

However, DHP estimates are dependent on the camera position. Downward looking overestimates the LAI up to 50% regarding Upward looking. Upward looking shows better consistency with LICOR estimates, with errors not larger than those found between LICORs estimates.

Concerning the spatial extension of in-situ measurements, the best results have been obtained using a transfer function derived from a Weighted Multiple linear regression.

Besides different sampling strategies were performed, the derived high-resolution LAI maps (LICOR and CAMERA) show relatively small differences (RMSE<0.3). Around 50 units covering the different crops seems to be enough for obtaining a good transfer function for the Barrax test site.


CONCLUSIONS

Comparison of In-Situ degradated maps with different products at 1km and 3km resolution over the small and large area shows that the UV LSA SAF algorithm on VEGETATION data shows the best correlation for FVC and LAI with an RMS < VGT_CYCL-V1< MODIS. MODIS FAPAR product shows a good agreement (RMS~0.08,r2~0.8). CYCLOPES products overestimate FVC and LAI, whilst underestimate FAPAR.

Open issues- Derive high-resolution maps for SPARC04- Assessment of the TF outside the study area (5x5km)- Evaluate different upscaling methods from 20m to 3km

Contact us for:-LAI/FVC/FAPAR data and maps [email protected] database and new ESA activities [email protected]

mailto:[email protected]

mailto:[email protected]


Thank you for your attention !!


Transfer Function FLAG IMAGE

Computation of convex hull over the collocated radiance values with the in situ measurements

Smallest convex region that contains the data set

Influence of Sampling Spatial Strategy

BIDIMENSIONAL COVEX HULL FOR THE BANDS COMBINATION (GREEN, RED, NIR)

RED-GREEN NIR-GREEN NIR-RED


LAI LICOR (0-5) FVC HP (0-1) fAPAR HP (0-1)AGGREGATION FOR VALIDATE LOW RESOLUTION

Barrax validation agricultural site: Leassons learnt during SPARC campaigns

Documents