Albedo estimation from PolDER data

Multi-spectral and multi-directional data acquired during the ReSeDA experiment thanks to the airborne PolDER sensor were used to retrieve surface albedo over the experimental site, for 16 days over the year 1997. The data were available in four wave-bands (10 or 20 nm width) centered at 443 nm, 550 nm, 670 nm, and 865 nm. Zenith view angles ranged from 0 to 50 o. This study aimed at evaluate a procedure based on the use of multi-directional and multi-spectral information to retrieve surface albedo. Multidirectional information was extracted thanks to BRDF kerneldriven models. We compared the performances of three models (Walthall, Roujean and MRPV) in the four PolDER channels. The spectrally integrated value of the albedo was then derived from the of the hemispherical reflectance estimates in the four wave-bands, thanks to the linear regressions proposed by Weiss et al. (1999). 20 m resolution albedo maps were computed, and then compared to field measurements over several crop fields considering all days of the experiment. Results showed that PolDER retrievals overestimated ground measurements. This might be explained, at least partially, by inappropriate linear combinations used for the spectral extrapolation.

1INRA Bioclimatologie, Domaine St Paul, 84914 Avignon Cedex 9, France

2CESBio, 18 avenue E.Belin, BP 2801, 31041 Toulouse Cedex 4, France

3CETP / IPSL / CNRS, 10-12 avenue de l’Europe, 78140 Velizy, France

Camera-ready Copy for

Physics and Chemistry of the Earth

Manuscript-No ???

Offset requests to:

F Jacob INRA-Bioclimatologie Domaine St Paul, Site Agroparc

84914 Avignon Cedex 9 France

Albedo estimation from PolDER data

F Jacob1, M Weiss1, A Olioso1, O Hautecoeur2,

C Franc¸ois3, M Leroy2, and C Ottl´e3

1INRA Bioclimatologie, Domaine St Paul, 84914 Avignon Cedex 9, France

2CESBio, 18 avenue E.Belin, BP 2801, 31041 Toulouse Cedex 4, France

3CETP / IPSL / CNRS, 10-12 avenue de l’Europe, 78140 Velizy, France

Received ??? – Accepted ???

Abstract Multi-spectral and multi-directional data acquired

during the ReSeDA experiment thanks to the airborne

Pol-DER sensor were used to retrieve surface albedo over the

experimental site, for 16 days over the year 1997 The data

were available in four wave-bands (10 or 20 nm width)

cen-tered at 443 nm, 550 nm, 670 nm, and 865 nm Zenith view

angles ranged from 0 to 50 o This study aimed at

eval-uate a procedure based on the use of multi-directional and

multi-spectral information to retrieve surface albedo

Multi-directional information was extracted thanks to BRDF

kernel-driven models We compared the performances of three

mod-els (Walthall, Roujean and MRPV) in the four PolDER

chan-nels The spectrally integrated value of the albedo was then

derived from the of the hemispherical reflectance estimates

in the four wave-bands, thanks to the linear regressions

pro-posed by Weiss et al (1999) 20 m resolution albedo maps

were computed, and then compared to field measurements

over several crop fields considering all days of the

experi-ment Results showed that PolDER retrievals overestimated

ground measurements This might be explained, at least

par-tially, by inappropriate linear combinations used for the

spec-tral extrapolation

1 Introduction

Surface albedo is defined as the fraction of incident solar

energy over the whole solar spectrum reflected in all

direc-tions (Pinty and Verstraete, 1992) It is especially important

for the global climate modeling (Dickinson, 1983), as well

as for surface fluxes estimation (Kustas et al., 1994; Olioso

et al., 1999) Generally, a relative accuracy of5% is

re-quired (Henderson-Sellers and Wilson, 1983)

In this study, we map surface albedo using multi-directional

and multi-spectral remotely sensed data acquired with the

airborne PolDER sensor during the ReSeDA experiment The

determination of albedo from remote sensing depends on two

Correspondence to: F JACOB

aspects: i) the anisotropic behavior of natural surfaces

re-quires the characterization of the angular distribution of the reflected solar radiation (expressed as BRDF for Bidirectional Reflectance Distribution Function) from the available

direc-tional measurements in a given wave-band; ii) the

determi-nation of the reflected energy over the whole solar spectrum from the wave-band estimates requires a spectral extrapola-tion Several methods have been developed to characterize the BRDF from satellite data Two classes may be distin-guished: the inversion of radiative transfer models, and the inversion of kernel-driven models As the first one is math-ematically complex and time consuming, we have chosen to consider the second one, which has been validated by several authors (see the review by Wanner et al (1997)) Generally, the spectral extrapolation is performed thanks to linear com-binations Several coefficient sets have been proposed and validated in the literature (e.g Tucker and Sellers (1986); Brest and Goward (1987); Song and Gao (1999)) In this study, we have chosen to use the coefficients proposed by Weiss et al (1999)

The ReSeDA experiment provided a framework with two interesting aspects for the validation of the proposed approach:

i) it covered the whole cycles of different types of crops

in-cluding winter (wheat) and summer crops (sunflower, corn);

ii) the high spatial resolution remote sensing data reduced

problems related to mixed pixels

2 Data acquisition and preprocessing

2.1 The ReSeDA Field Experiment

The ReSeDA experiment lasted from December 1996 to December 1997, in the South East of France (N 43o47’, E 4o 45’) The experimental site was a small agricultural region (55 km2

) with sunflower, wheat, corn, grassland and alfalfa fields with a mean size of 200200 m2

(Pr´evot et al., 1998; Olioso et al., 1998)

Airborne PolDER data were acquired approximately one

or two times per month, on clear sky days and at a 3000 m

altitude involving a 20 m nadir spatial resolution Four flight

lines were parallel to the solar plan, and one perpendicular

These five lines were completed within 45 minutes centered

at the solar noon The data were available in four wave-bands

(10 or 20 nm width) centered at 443 nm, 550 nm, 670 nm,

and 865 nm Zenith view angles ranged from 0 to 50o

Sensor calibration was performed by the L.O.A

(Labora-toire d’Optique Atmosph´erique, Lille, France) with a 3 month

frequency The procedure accounted for ambient

tempera-ture, dark current, and inter-calibration of CCD matrix

de-tector Its accuracy was about 5%

Atmospheric effects were corrected thanks to the SMAC

algorithm (Rahman and Dedieu, 1994) based on the

inver-sion of the atmospheric radiative transfer model 6S (Vermote

et al., 1997) The required information consisted in aerosol

optical thickness, water vapor content, both estimated from

field sunphotometer measurements, and ozone atmospheric

concentration obtained from TOMS climatic daily data

Each image was registered thanks to a Global Positioning

System and an inertial central data, according to a Lambert II

projection This projection provided a spatial sampling of the

site corresponding to a grid of 250250 pixels with a 20 m

resolution

All these pre-processing are described in details by Leroy

et al (2000) They allowed to derive BRDF samplings that

depended on both the location on the site and the flight line

configuration

2.3 Field data

Field measurements of albedo were performed on seven

locations corresponding to alfalfa, wheat, and sunflower crops

Albedo was deduced from measurements of incident

radia-tion using a Kipp pyranometer located on the

meteorologi-cal site, and measurements of reflected radiation using Kipp

pyranometers or Skye silicon sensors looking to the ground

surface The data set corresponded to 20 minutes mean

val-ues having a circular footprint between 1000 and 3000 m2

The spectral ranges of Kipp and Skye sensors were

differ-ent, corresponding respectively to [300-3000] nm and

[400-1100] nm For the latter, it was necessary to consider the

spectral behavior of the observed surface, in order to

ex-trapolate the estimates over the whole solar spectrum This

has been performed thanks to a formulation of the actual

albedo as a function of the measured one The

formula-tion was calibrated over simulaformula-tions of the radiative transfer

model SAIL (Verhoef, 1984, 1985) performed by Franc¸ois

et al (2000) Model input variables were soil and leaf

op-tical properties, incident solar radiation from simulations of

the atmospheric radiative transfer model 6S (Vermote et al.,

1997) that took account for numerous atmospheric situations,

and measurements of Leaf Area Index (LAI) Simulations of

actual albedo and Skye estimates are described in details by

ear shape, inducing a residual error of 0.003:

Albedo actual

= 0:781  Albedo

Sk ye

3.1 Position of the problem From the definition given in Sect 1, the instantaneous albedo

a(

s

; '

s is expressed as following (

sand'

sare respecti-vely the solar zenith and azimuth angles):

a(

s

; ' s

=

3000nm R

300nm

 h;

(

s

; ' s R g;

d

3000nm R

300nm R g;

d

(2)

where is the wavelength The spectral albedo or hemi-spherical reflectance

h;

(

s

; '

s represents the fraction of the spectral incoming solar radiationR

g; reflected in the whole hemisphere It is expressed through the bidirectional reflectance

 (

s

; ' s

;  v

; '

v (

vand'

vare respectively the view zenith and azimuth angles):

 h;

(

s

; ' s

=

2

R

0

=2 R

0



 (

s

; ' s

;  v

; ' v cos  v sin  v d

v d' v

(3)

PolDER provided measurements of bidirectional reflec-tances

 (

s

; ' s

;  v

; '

v in the four considered channels From these directional samplings, we estimated hemispherical re-flectances

h;

(

s

; '

s by inverting BRDF kernel-driven mod-els, and then the instantaneous albedo using a simple spectral extrapolation procedure Both aspects are presented below 3.2 Retrieving hemispherical reflectance using BRDF kernel-driven models

The philosophy of a BRDF kernel-driven model is to ex-press the bidirectional reflectance

 (

s

;  v

; ' s

; '

v thanks

to a linear combination ofnkernelsN

i(a kernel is a prede-fined function of view and solar angles):



 (

s

;  v

; ' s

; ' v

= n X

i=1 i;

N i (

s

;  v

; ' s

; '

where

i;are the weighting coefficients The number and the formulation of the kernelsN

idiffer from one model to another Among the large number of kernel-driven models that were developed these last years, we have chosen to test three of them: Walthall (Walthall et al., 1985), Roujean (Rou-jean et al., 1992) and MRPV (Engelsen et al., 1996) Seve-ral studies showed that MRPV was the most accurate model both for the accuracy of the fitting and the extrapolation capa-bilities, while Walthall and Roujean were often presented as robust models (Baret et al., 1997; Wanner et al., 1997; Weiss

Coefficient Blue Green Red Near Infra-Red

Set (445 nm) (560 nm) (665 nm) (865 nm)

Set n o

Set n o

Set n o

Table 1 Sets of coefficients for the computation of the albedo as a linear

combination of wave-band hemispherical reflectances.

et al., 2000) We should notice that Roujean and Walthall are

linear models, while MRPV is a semi-linear one

The weighting coefficients

i; might be obtained by in-verting the model from the multi-angular data set This was

performed for each pixel and each PolDER channel using

the procedure described by Weiss et al (2000) The retrieved

BRDF through these coefficients were then integrated to

ob-tain the hemispherical reflectances

h;j in the PolDER chan-nels (j = 1; : : 4)

3.3 Spectral extrapolation from PolDER channels

The spectral extrapolation was based on the assumption

that for a given wavelength 2 [300 3000]nm, the

hemi-spherical reflectance

h;

(

s

; '

s is a linear combination of the hemispherical reflectance 

h;j (

s

; '

s estimated in the four channels PolDER Then, it was possible to express the

albedo as:

a(

s

; '

s

=

4

X

j=1

j :

h;j (

s

; '

Several studies have been devoted to the determination of

the coefficients

j, but there is not at the present time any proposition for the PolDER sensor In this context, we have

chosen to test three sets of coefficients proposed by Weiss

et al (1999) when considering blue, green, red and near

in-frared channels (corresponding to 445, 560, 665 and 865

nm) These coefficients were obtained from a linear

regres-sion calculated over numerous soil coverage situations by the

radiative transfer model DISORD (Myneni et al., 1992),

be-tween 400 and 2500 nm These simulations were

represen-tative of several kinds of canopies at three different latitudes

and for three days corresponding to different seasons

Coeffi-cient sets are given in Table 1 The relative accuracy of these

linear regressions was estimated as the Root Mean Square

Er-ror (RMSE) between simulated and retrieved albedo: it was

about 7%

4 Results and validation

4.1 Performances of BRDF kernel-driven models

We evaluated the BRDF retrieval performances of the

kernel-driven models by calculating for each pixel the absolute RMSE

and the relative RMSE (RRMSE) between observed ( )

RMSE 0.0138 0.0118 0.0129 0.0280 MRPV

RMSE 0.0112 0.0123 0.0138 0.0290 Roujean

RMSE 0.0116 0.0136 0.0150 0.0303 Walthall

Table 2. Absolute and relative RMSE between observed and retrieved BRDF through the three kernel-driven models for the 10 April 1997.

and retrieved (

ret j

) bidirectional reflectances:

v

u

t

k =m P

k =1 (

obs j

ret j (k)) 2

m

(6)

<  obs j

>

(7)

where< 

obs j

>is the mean value of themobserved bidi-rectional reflectances Results showed that the greatest er-rors occurred for pixels located on both the Alpilles moun-tain chain and field edges In the first case, this might be explained by the inadequacy of BRDF models when they are applied to mountainous areas In the second case, it might

be explained by the combination of registration inaccuracy and spatial variability Table 2 presents the RMSE and the RRMSE over the whole site for a representative day The er-rors in the blue channel were more important whatever was the model, and maybe induced by the perturbations occur-ring in this channel such as the inaccuracy of the sensor cal-ibration or the residual noise due to atmospheric diffusion

by aerosols The BRDF retrieval performances of the three models were very similar and slightly better for MRPV at

550, 670 and 865 nm However, this model presented a great sensitivity to the perturbations mentioned previously This high sensitivity could be explained by the semi-linear model formulation When considering BRDF retrieval per-formances without pixels located on both the mountain chain and field edges, the models gave slightly better results, with

a lower RRMSE about 3 to 5% On the other hand, these performances were quite better when considering only pix-els located on the field measurements, with a RRMSE di-vided by 2 We explained this by the homogeneity around field measurements locations, inducing small perturbations due to the combination of image registration inaccuracy and spatial variability

The comparison of the hemispherical reflectance estimates over the whole site showed differences between MRPV and the two others models These differences were more im-portant in the blue channel, for which MRPV provided nu-merous unrealistic values such as hemispherical reflectances close to one Moreover, we observed that differences be-tween models decreased with respect to the wave-band

There-0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6

0.2

0.25

0.3

0.35

0.4

0.45

0.5

0.55

0.6

ρh from Roujean model

ρ h

Fig 1 Comparison of hemispherical reflectance estimates from Walthall’s

and Roujean’s models for the channel 865 nm when considering pixels

lo-cated on field measurements The solid line represents the linear regression

between the estimates from the two models.

fore, the hemispherical reflectance retrieval through

kernel-driven BRDF models should be more stable with an increase

of the wavelength This observation was in agreement with

the conclusions of Baret et al (1997) When considering

only pixels located on field measurements, Roujean

over-estimated the hemispherical reflectance as compared to the

others models (see an example with Fig 1), while the

under-estimation was observed for MRPV at 443 and 670 nm, and

for Walthall at 550 and 865 nm

4.2 Validation of albedo estimates

Albedo calculations have been performed considering the

three BRDF kernel-driven models and the three sets of

coeffi-cients Therefore, nine albedo maps were computed for each

day of the experiment (see for example Fig 2) These maps

depicted albedo values between 0.1 and 0.4 This important

variability was explained by the simultaneous presence on

the site of vegetative surfaces and bare soils As expected,

the lowest values corresponded to well vegetated fields or

wet bare soils, while the highest ones corresponded to dry

bare soils or very sparse vegetation

Since the field data and PolDER pixels had different

foot-prints (Sect 2), we assessed the impact of the spatial

variabil-ity on the airborne albedo estimates by computing the relative

standard deviation (standard deviation / mean value) inside

both 33 and 55 PolDER pixel windows The results,

be-tween 1 and 2%, underlined the negligible effect of the

spa-tial variability around field measurement locations as much

as the window size was smaller than the field one Therefore,

we decided to perform the validation by extracting PolDER

estimates through 33 pixels windows

An example of comparison between field and airborne

es-timates of the albedo is given in Fig.3 for one kernel-driven

0.15 0.2 0.25 0.3

Fig 2 Albedo map for the 29 July 1997 using the MRPV model and the

coefficient set n o

3 The Alpilles mountain chain has been removed.

model and one set of coefficients These comparisons showed that were no differences between Kipp and Skye estimates af-ter the corrections of the lataf-ter (Sect.2) For each of the nine possibilities, we computed the absolute RMSE as in eq.6 and relative RMSE (RRMSE) as in eq.7, as well as the absolute bias (Bias) and relative bias (RBias) calculated as:

k =M P

k =1 (a PolDE

in situ (k))

M

(8)

Bias

< a

in situ

>

(9) wherea

olDE

is the albedo estimated from PolDER data,

a

in situ

is the albedo measured in-situ, and< a

in situ

>

is the mean value of theMfield data The results are given

in Table 3 Airborne retrievals were systematically higher than field estimates Considering each model, an important overestimation occurred with the first coefficient set that

cor-BRDF Model

& Coefficient set RMSE RRMSE Bias RBias MRPV & set 1 0.0530 27.2% 0.0480 24.7% MRPV & set 2 0.0345 17.7% 0.0285 14.7% MRPV & set 3 0.0320 16.4% 0.0251 12.8% Roujean & set 1 0.0532 27.7% 0.0504 26.2% Roujean & set 2 0.0378 19.7% 0.0333 17.3% Roujean & set 3 0.0348 18.1% 0.0301 15.6% Walthall & set 1 0.0428 22.2% 0.0387 20.1% Walthall & set 2 0.0280 14.6% 0.0217 11.3% Walthall & set 3 0.0255 13.2% 0.0183 9.5%

Table 3 Absolute and relative RMSE and Bias between airborne and field

estimates of the albedo The solid line represents the linear regression

be-tween the in-situ and airborne estimates.

0.1 0.15 0.2 0.25 0.3

0.1

0.12

0.14

0.16

0.18

0.2

0.22

0.24

0.26

0.28

0.3

In−situ estimates

Field estimates from Kipp sensors Field estimates from Skye sensors

Fig 3 Comparison between field and airborne albedo for the whole

Re-SeDA experiment, considering the MRPV model and the coefficient set n 0

3.

responded to the contributions of the red and NIR channels,

while this overestimation decreased with a decrease in the

red and NIR channels contributions The comparison from a

model to another with the same set of coefficient showed that

the highest estimates were obtained with Roujean, while the

lowest ones corresponded to Walthall These observations

were explained as following:

– Roujean provided the highest hemispherical reflectances

whatever was the channel, and therefore the highest albedo

values;

– since the absolute value of the bias between Walthall

and MRPV hemispherical reflectances was lower in the

red (0.0004) than in the NIR (-0.0125), Walthall yielded

the lowest albedo values with the set of coefficient no

1 ;

– Walthall provided the lowest albedo values with the sets

of coefficients no

2 and no

3 because it yielded the lowest hemispherical reflectances in the green and NIR

chan-nels

These results showed that the albedo retrieval strongly

de-pends on the hemispherical estimates, and then requires

ac-curate ones Besides, the RMSE and bias were generally

close from a model to another, while the high RMSE

val-ues were partially induced by the bias At the present time,

we think that this general overestimation could result from

either the hemispherical reflectance estimates or the

assump-tions used when calibrating the linear combination (Sect.3)

Indeed, the simulations performed by Weiss et al (1999) to

estimate the coefficients corresponded to the spectral

inter-val [400 - 2500] nm, while the whole solar spectrum ranges

between 300 and 3000 nm Therefore, the incident solar

radi-ation were lower than the actual one, by 6-8% referring to the

works of Avaste et al (1962) At a lower extent, the accuracy

of these sets of coefficients was affected by the spectral

dif-ference between the radiative transfer model simulations and

BRDF Model

& Coefficient set a b RMSEU RRMSE U MRPV & set 1 0.911 0.065 0.0223 9.2% MRPV & set 2 0.810 0.065 0.0185 8.3% MRPV & set 3 0.818 0.061 0.0190 8.7% Roujean & set 1 0.919 0.066 0.0169 7.0% Roujean & set 2 0.804 0.071 0.0167 7.4% Roujean & set 3 0.816 0.065 0.0164 7.4% Walthall & set 1 0.986 0.041 0.0182 7.9% Walthall & set 2 0.866 0.048 0.0172 8.0% Walthall & set 3 0.879 0.042 0.0174 8.2%

Table 4 Coefficient of the linear regression between field and airborne

es-timates of the albedo (a: slope, b: offset), and RMSE between PolDER estimate and the linear regression (RMSE U ).

the PolDER wave-bands, as well as by the residual noises due to instrumental and atmospheric effects

In order to assess the accuracy that it would be possible

to achieve, we calculated the coefficients of the linear

re-gression between predicted (or airborne) and observed (or

in-situ) estimates, as well as the absolute and relative

unsystem-atic RMSE (RMSEU and RRMSEU) (see the Table 4) The RMSEUcomputes the scattering around the linear regression

as the RMSE between the predicted values corrected from this regression and the actual ones (Kustas et al., 1996) The coefficients of the linear regression suggested that consider-ing only hemispherical reflectances in red and NIR induced mainly an offset, while using more wave-bands seemed to provide an overestimation of low albedo values and an under-estimation of high ones Finally, the RMSEU computations showed that it would be possible to achieve an absolute ac-curacy between 0.0164 and 0.0223 for albedo values ranging from 0.1 to 0.25 after the removal of slopes and offsets This would correspond to a relative accuracy ranging between 7 and 9% We should notice that in this case, the result cor-responding to the lower discrepancy would be obtained with the Roujean model when considering the hemispherical re-flectances in the red and the NIR channels

5 Conclusions

The objective of this study was to map albedo on the Re-SeDA experiment site, using the airborne multi-spectral and multi-directional Vis-Near Infra-Red PolDER remote sens-ing data Moreover, these high spatial resolution and multi-temporal data allowed to perform a validation with less prob-lems related to mixed pixels and over cycles of several crops The multi-directional information was extracted through BRDF kernel-driven models We tested three models (MRPV, Roujean and Walthall) that gave similar results for both the BRDF retrieval and the hemispherical reflectance estimation However, results showed that the data set acquired in the blue channel have to be considered with care; and that the MRPV

ric processing and image registration.

The multi-spectral information was used by computing the

albedo as a linear combination of the hemispherical

reflec-tances in PolDER channels We tested three sets of

coeffi-cients previously proposed as generic ones by Weiss et al

(1999)

The validation thanks to field measurements underlined an

overestimation whatever were the BRDF model and the set

of coefficients This could be explained by either the

hemi-spherical reflectance estimates or the assumptions used for

the calibration of the linear combination The first point

could results from the PolDER measurements or the

kernel-driven model retrievals The second point could result from

the underestimation of the incident solar radiation This could

be improved by considering the whole solar spectrum when

calibrating the linear regression We observed that the

re-moval of this overestimation should yield an absolute

accu-racy between 0.0164 and 0.0223 for albedo values ranging

from 0.1 to 0.25 (which corresponds to a relative accuracy

between 7 and 9%) Another possibility could be to take into

account surface properties through the NDVI when

calibrat-ing the linear combination, as proposed by Song and Gao

(1999)

In the future, these maps could be used as a reference for

validation at larger scale considering sensors such as NOAA

/ AVHRR for instance, and as inputs for surface energy

bal-ance calculation models (Jacob et al., 2000) However, one

should note that for pixels far from the site center, the

Pol-DER directional sampling quality was very poor and

there-fore that these results must be considered with care

References

Avaste, O., Moldau, H., and Shifrin, K., Distribution spectrale des

rayon-nements directs et diffus, Instrumental and Physical Astronomy, 3, 44–

57, 1962.

Baret, F., Weiss, M., Leroy, M., Hautecoeur, O., Santer, R., and B´egu´e,

A., Impact of surface anisotropies on the observation of optical

imag-ing sensors, final report, Esa contract 11341/95/nl/cn, ESA, ESTEC, the

Netherlands, 1997.

Brest, C and Goward, S., Deriving surface albedo measurements from

nar-row band satellite data, International Journal of Remote Sensing, 8, 351–

367, 1987.

Dickinson, R., Land surface processes and climate-surface albedos and

en-ergy balance, Advances in Geophysics, 25, 305–353, 1983.

Engelsen, O., Pinty, B., Verstraete, M., and Martonchik, J., Parametric

bidi-rectional reflectance factor models : evaluation, improvements and

ap-plications, Report eur16426en, European Commission, Joint Researches

Center, Space Application Institute, ISPRA, Italy, 1996.

Franc¸ois, C., Ottl´e, C., and Olioso, A., Correction of silicon sensors albedo

measurements using a canopy radiative transfer model, in Physics and

Chemistry of the Earth, EGS symposium, special ReSeDA session,

sub-mitted, 2000.

Henderson-Sellers, A and Wilson, M., Surface albedo data for climatic

modelling, Reviews on Geophysics, 23, 1743–1778, 1983.

Jacob, F., Olioso, A., Gu, X., Hanocq, O., Hautecoeur, O., and Leroy,

M., Mapping surface fluxes using Visible - Near Infra-Red and Thermal

the Earth, EGS symposium, special ReSeDA session, submitted, 2000.

Kustas, W., Moran, M., Humes, K., Stannard, D., Pinter, P., Hipps, L., Swiatek, E., and Goodrich, D., Surface energy balance estimates at lo-cal and regional slo-cales using optilo-cal remote sensing from an aircraft

plat-form and atmospheric data collected over semiarid rangelands, Water Re-sources Research, 30, 1241–1259, 1994.

Kustas, W., Humes, K., Norman, J., and Moran, M., Single and dual source modeling of surface energy fluxes with radiometric surface temperature,

Journal of Applied Meteorology, 35, 110–121, 1996.

Leroy, M., Hautecoeur, O., Berthelot, B., and Gu, X., The airborne polder

data during the reseda experiment, in Physics and Chemistry of the Earth, EGS symposium, special ReSeDA session, submitted, 2000.

Myneni, R., Asrar, G., and Hall, F., A three-dimensional radiative transfer

method for optical remote sensing of vegetated land surfaces, Remote Sensing of Environment, 41, 105–121, 1992.

Olioso, A., Pr´evot, L., Baret, F., Chanzy, A., and et al., Spatial aspects in

the alpilles-reseda project, in Scaling and modeling in forestry: applica-tion in remote sensing and GIS, Ed D.Marceau, Universit´e de Montr´eal, Qu´ebec, pp 92–102, 1998.

Olioso, A., Chauki, H., Courault, D., and Wigneron, J., Estimation of evap-otranspiration and photosynthesis by assimilation of remote sensing data

into svat models, Remote Sensing of Environment, 68, 341–356, 1999.

Pinty, B and Verstraete, M., On the design and validation of surface

bidi-rectional reflectance and albedo model, Remote Sensing of Environment,

41, 155–167, 1992.

Pr´evot, L., Baret, F., Chanzy, A., Olioso, A., and et al., Assimilation of multi-sensor and multi-temporal remote sensing data to monitor

vegeta-tion and soil: the Alpilles ReSeDA project, in IGARSS’98 (Seattle, WA, USA), International Geoscience and Remote Sensing Symposium, Ed L Tsang, pp 17–30, 1998.

Rahman, H and Dedieu, G., Smac : a simplified method for the atmospheric

correction of satellite measurements in the solar spectrum, International Journal of Remote Sensing, 16, 123–143, 1994.

Roujean, J.-L., Leroy, M., and Deschamps, P., A bidirectional reflectance model of the earth’s surface for the correction of remote sensing data,

Journal of Geophysical Research, 97, 20 455–20 468, 1992.

Song, J and Gao, W., An improved method to derive surface albedo from narrowband avhrr satellite data: narrowband to broadband conversion,

Journal of Applied Meteorology, 38, 239–249, 1999.

Tucker, C and Sellers, P., Satellite remote sensing of primary production,

International Journal of Remote Sensing, 7, 1395–1416, 1986.

Verhoef, W., Light scattering by leaf layers with application to canopy

re-flectance modeling : the sail model, Remote Sensing of Environment, 16,

125–141, 1984.

Verhoef, W., Earth observation modelling based on layer scattering

matri-ces., Remote Sensing of Environment, 17, 165–178, 1985.

Vermote, E., Tanr´e, D., Deuz´e, J., and Morcrette, J., Second simulation of

the satellite signal in the solar spectrum: an overview, IEEE Transactions

on Geosciences and Remote Sensing, 35, 675–686, 1997.

Walthall, C., Norman, J., Welles, G., Campbell, G., and Blad, G., Sim-ple equation to approximate the bidirectional reflectance from vegetative

canopies and bare soil surfaces, Applied Optics, 24, 383–387, 1985.

Wanner, W., Strahler, A., Hu, B., Lewis, P., Muller, J.-P., Li, X., Barker Schaaf, C., and Barnsley, M., Global retrieval of bidirectional re-flectance and albedo over land from EOS MODIS and MISR data: theory

and algorithm., Journal of Geophysical Research, 102, 17 143–17 161,

1997.

Weiss, M., Baret, F., Leroy, M., B´egu´e, A., Hautecoeur, O., and Santer, R., Hemispherical reflectance and albedo estimate from the accumulation

of across-track sun-synchronous satellite data, Journal of Geophysical Research, 104, 22 221–22 232, 1999.

Weiss, M., Jacob, F., Baret, F., Pragn`ere, A., Leroy, M., Hautecoeur, O., Pr´evot, L., and Bruguier, N., Evaluation of kernel-driven brdf models for

the normalization of alpilles/reseda polder data, in Physics and Chemistry

of the Earth, EGS symposium, special ReSeDA session, 2000.

olDE

is the albedo estimated from PolDER data,

a

in situ

is the albedo measured in-situ, and<... first point

could results from the PolDER measurements or the

kernel-driven model retrievals The second point could result from

the underestimation of the incident solar... 1), while the

under -estimation was observed for MRPV at 443 and 670 nm, and

for Walthall at 550 and 865 nm

4.2 Validation of albedo estimates

Albedo calculations have