611

Agronomie 22 (2002) 611–618 © INRA, EDP Sciences, 2002 DOI: C. 10.1051/agro:2002033 Conversion François of et al. 400–1100 nm vegetation albedo

Original article

Conversion of 400–1100 nm vegetation albedo measurements into total shortwave broadband albedo using a canopy radiative transfer model Christophe FRANÇOISa*, Catherine OTTLÉa, Albert OLIOSOb, Laurent PRÉVOTb, Nadine BRUGUIERb, Yannick DUCROSb aCentre

d’Études des Environnements Terrestre et Planétaires, 78140 Vélizy, France bINRA-Bioclimatologie, 84914 Avignon, France (Received 30 August 2001; accepted 9 April 2002)

Abstract – The surface albedo was measured during the ReSeDA experiment over several fields with two different devices: Kipp thermopiles and Skye silicon photopiles. The comparison of different datasets shows a systematic bias between the measurements performed by the two devices: the albedos measured by the Skye instruments are overestimated compared to the Kipp ones. This difference is due to a specific problem with the Skye radiometers, which were designed for incident solar radiation measurements, whereas they were used to measure reflected crop radiation. Moreover, the Skye filter function uses a bandwidth corresponding to the 400–1100 nm region only, and not the whole spectra. The SAIL model was used to correct the Skye measurements by simulating the response of the instrument and the true albedo on each experimental field. After this correction, the albedos actually show a good agreement between the data acquired by the two devices. albedo / vegetation / thermopile / photopile / radiative transfer model Résumé – Utilisation d’un modèle de transfert radiatif pour la conversion de mesures d’albédos 400–1100 nm en albédo total. Lors de l’expérience ReSeDA l’albédo de surface a été mesuré avec deux appareils différents : des thermopiles Kipp et des photodiodes au silicium Skye. La comparaison des différents jeux de données a montré un biais systématique entre les deux types de mesures : les albedos mesurés par les instruments Skye sont surestimés par rapport à ceux mesurés avec les thermopiles Kipp. La différence est due à un problème spécifique aux radiomètres Skye qui ont été conçus pour mesurer le rayonnement solaire incident, mais qui ont été utilisés pour mesurer le rayonnement réfléchi par le couvert. De plus la fonction filtre des instruments Skye n’utilise pas le spectre complet mais une région limitée au domaine 400–1100 nm. Le modèle SAIL a été utilisé pour corriger les mesures Skye en simulant la réponse des instruments Skye et l’albédo vrai pour chaque parcelle expérimentale. Après cette correction, les albédos obtenus à partir des deux appareils montrent un bon accord. albedo / végétation / thermopile / photopile / modèle de transfert radiatif

1. INTRODUCTION The main objective of the ReSeDA project is the use of multisensor and multitemporal observations for the monitoring of soil and vegetation processes, on local and regional scales, including assimilation of remote sensing data into canopy and soil functioning models. The final goal is to propose methods to estimate net primary production,

evapotranspiration and yield of cultivated crops. In the context of surface energy balance applications, albedo is a parameter of great importance. The required accuracy on albedo measurements varies from one application to another. An absolute accuracy about ± 2% is a common requirement [7]. Albedo measurements were performed during the ReSeDA experiment over different fields, mainly wheat and

Communicated by Thomas Schmugge (Beltsville, USA) * Correspondence and reprints [email protected] Present address: Écologie, Systématique et Évolution, Bât 362, Dpt. Écophysiologie Végétale, 91405 Orsay Cedex, France.

612

C. François et al.

sunflower [5, 6]. Classical Kipp thermopiles have been used for three of these fields, and Skye silicon photopiles were used for four of these fields. The comparison of both datasets exhibited a systematic difference between the measurements performed by the two devices. This difference is due to the calibration of the Skye radiometers, which were calibrated for incident solar radiation measurements, and then used to measure reflected crop radiation, and to the filter function of the Skye radiometers (which does not cover the whole solar spectrum but a bandwidth corresponding to 400–1100 nm only). A correction of the Skye albedo measurements was therefore desirable to obtain consistent albedo measurements over all fields during the ReSeDA experiment; unfortunately, no coincident measurements of Kipp and Skye albedos were made. Instead, we used a radiative transfer model for horizontal homogeneous canopies (SAIL, [9]) to simulate albedos: total shortwave apparent albedo (as defined in [4]), hereafter referred to as true albedo, and Skye albedo (with the Skye filter function) were simulated to study the differences and derive a correction. The choice of a simple model like SAIL was made because the considered crops (wheat, maize, sunflowers) do not exhibit atypical behavior or large organs. Thus the validity of SAIL was assumed. For the special case of the sunflower crop, only corrections made for small capitulum were kept.

2. MATERIALS AND METHODS

The reflected radiation was measured either by Kipp pyranometer or a Skye photopile, depending on the field. The filter function of the Skye silicon photopiles is highly variable with the wavelength and limited to a small spectral range only, unlike the Kipp radiometer (see Fig. 1). Figure 2 shows the variation of the albedo for two wheat fields (field 101 – where a Kipp radiometer was used, and field 120 – where a Skye radiometer was used) with the same characteristics throughout the year (variety of wheat, sowing date). Average albedos (between 10 a.m. and 2 p.m.) are presented in this graph. The problem is an overestimation of the albedo measured with the Skye radiometer compared to the Kipp one. Both albedos were not expected to be exactly equal since two different fields were observed, but the observed systematic difference (greater than 0.05 and reaching 0.08) was believed to be excessive given the similarity between the two fields (see the comparison of LAI between the two fields in Fig. 3). A look at both Figures 2 and 3 reveals that on the field 101 the albedo is increased by about 0.05 when the LAI increases from 1 to 2. Because of the nonlinear relationship between LAI and albedo, an increase in LAI from 2 to 3 (the maximum LAI difference between field 101 and 120) would correspond to an increase smaller than 0.05 in albedo (around 0.03 according to SAIL simulations in the field conditions), whereas the observed maximum difference between both instruments on the two fields is much higher: 0.08. So, the SAIL radiative transfer model was used to study the differences between Skye and true albedos, and eventually to correct the Skye albedos.

2.1. Albedo measurements Albedo determination involves two measurements: the incident global radiation and the radiation reflected by the canopy (covering 320–3000 nm). The incident global radiation was measured using three Kipp pyranometers in three different plots (very few variations were observed between plots).

2.2. The SAIL model The SAIL model [9] is usually used for directional reflectance calculations. The albedo a is an average of the spectral albedos ai calculated here with a bandwidth Dl = 0.20 mm, and weighted by a device filter function fi if

Figure 1. Normalized filter function of the Skye (silicon sensor) and Kipp (thermopile) instruments in function of the wavelength (nm).

613

Conversion of 400–1100 nm vegetation albedo 5

necessary (f is flat for true albedos, or follows the Skye filter function presented in Fig. 1 for Skye albedos):

field 101 field 120

4.5

0. 32 – 2. 50 mm

a=

∑a

i

fi ∆ l

(1)

4

i

3.5

0. 32 – 2. 50 mm

with

∑

f i ∆ l = 1 ( ∆ l = 0.20 mm).

3

The chosen spectral domain for the simulation (320–2500 nm) is shorter than the total shortwave region (320–3000 nm), but the incident radiation, as well as the leaf reflectance, is negligible after 2500 nm. The spectral albedos ai are calculated from the fractions of direct (es) and diffuse (ed) incident solar radiation, and from the direct (propsi) and diffuse (propdi) normalized incident solar spectra. The coefficients propsi and propdi (mm–1) allow us to transform the integrated incident global radiation into spectral radiations, for both the direct and diffuse parts. No bidirectional reflectances were used in this study: the SAIL model was used to compute the direct to diffuse (s → d) and diffuse to diffuse (d →d) hemispherical reflectance components Rsd and Rdd, and thus the spectral albedos: ( R ) e props i + ( Rdd ) ed propd i (2) a i = sd s es props i + e d propd i with

0. 32 – 2. 50 mm

0. 32 – 2. 50 mm

i

i

∑ props i ∆ l = 1 and

∑ propd ∆ i

l

LAI

i

2.5

2

1.5

1

0.5

0

60

80

100

120

140

160

Day of Year 1997

Figure 3. Comparison of the LAI measurements on the two wheat fields (field 101 and field 120).

= 1(3)

and es + ed = 1. (4) During the ReSeDA experiment, the fractions of direct and diffuse radiation (es and ed) were measured. It should be noted that Rsd and Rdd (i.e. coefficients propsi and propdi) have to be calculated at the Top of Canopy (TOC) and not at the Top of Atmosphere (TOA). While the TOA normalized incident spectrum may be considered as constant, the TOC incident spectrum is much more variable, due to atmospheric conditions (aerosol loading, water vapor content) and view angle. This problem was solved by performing a wide range of simulations using the 6S atmo-

Figure 4. Scatter plot showing the distribution of water vapor contents (W0) for the 485 non-cloudy maritime radio soundings extracted from a representative global database (TIGR, [1]). The chosen range (0.5 to 3.5 g/cm2) encompasses most mid-latitude encountered situations.

1 Field 101 Wheat (Kipp) Field 120 Wheat (Skye)

0.9

Albedo (10h - 14h)

0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

Jan

Feb

Mar

Apr

May

Jun

Jul

Aug

Sep

Oct

1997

Figure 2. Comparison of albedo measurements (averaged between 10 a.m. and 2 p.m.) on two wheat fields, using two different devices: Kipp radiometers on field 101 and Skye radiometers on field 120. The high albedo peak around day 360 (December-January) shows the presence of snow over the fields.

C. François et al.

spheric radiative transfer model [10] for a variety of aerosol optical thicknesses (AOT) at 550 nm (from 0.2 to 0.5, corresponding to a visibility ranging from 8.5 to 47.7 km), water vapor content (from 0.5 to 3.5 g/cm2, corresponding to most situations encountered in mid-latitudes, Fig. 4) and solar incident angle qs (air mass m = 1 to air mass m = 5, corresponding to solar zenith angles ranging from 0o (nadir) to 78.5o). 6S was modified to obtain the corresponding diffuse and direct TOC solar incident spectra. These spectra were normalized (Eq. (3)) and the resulting diffuse and direct TOC solar incident spectra (in µm–1) are illustrated in Figure 5: note the two spectra are different, each one with a characteristic pattern. Variations of the spectra for the different atmospheric conditions after normalization were considered sufficiently small and acceptable for our purpose. In this study four classes of spectra were averaged (m = 1, m = 1.5, m =2 and m = 2.5, i.e. qs=0o to qs=66o). These averages were calculated for each case (direct or diffuse) and used as a “universal” normalized incident solar radiation spectrum for SAIL. We obtained in this manner propsi and propdi with a 20 nm step (from 320 nm to 2500 nm). It must be noted that these spectra are valid for a cloud-free atmosphere only, since no clouds were introduced in the 6S simulations. The diffuse radiation arises therefore only from water vapor and aerosols in the atmosphere, with standard values for ozone, CO2 and other atmospheric constituents.

160 140

Number of situations

614

120 100 80 60 40 20 0

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Diffuse radiation fraction e d

Figure 6. Histogram showing the distribution of diffuse radiation fraction (ed). The bimodal distribution corresponds to clear sky situations (left) and overcast situations (right). The threshold for clouds has been fixed at ed =0.5.

A threshold on ed for cloudy sky conditions has to be determined: when the sky is overcast by clouds, the measured “diffuse” fraction of solar incident radiation ed is 1; when the sky is clear, ed is around 0.23 [8]. The distribution of ed for the months of interest (May to September) was plotted (Fig. 6) considering data between 10 a.m. and 2 p.m. (according to our correction scheme). We observed a bimodal distribution and we chose the threshold ed=0.5 (i.e. 50% diffuse radiation) as the end of the clear sky peak. As a consequence, no simulations were performed when the measured “diffuse” fraction of solar incident radiation ed was greater than 0.5, assuming that the simulations and/or corrections would have been incorrect due to the cloud impacts. 2.3 Correction method The SAIL model is used to simulate both true (i.e. Kipp) and Skye albedos. The measured Skye albedo is then corrected using the following ratio: corrected albedo =

true ( simulated ) albedo SKYE measurements. SKYE ( simulated ) albedo

(5) While the simulations may not be accurate relative to actual data, the ratio of the simulations has to be realistic. 2.4 Sensitivity study Figure 5. Top of canopy direct and diffuse radiation solar spectra (in mm–1) for different solar zenith angles qs. The solar angles are expressed in air mass (m = 1/cosqs): m = 1, qs=0o; m = 2, qs=60o; m = 3, qs=72.5o; m = 4, qs=75.5o. The spectra used in this study are averaged between m=1 (qs=0o) and m = 2.5 (qs=66o).

A sensitivity study was performed: the variation of the ratio was studied when the input parameters were changed. The input parameters include the leaf reflectance spectrum, soil reflectance spectrum, Leaf Inclination Distribution Function

Conversion of 400–1100 nm vegetation albedo

(LIDF), Leaf Area Index (LAI), solar inclination and fraction of diffuse radiation ed. In order to study the influence of the leaf reflectance spectrum, the SAIL model was coupled with a leaf radiative transfer model, PROSPECT [3], which simulates the leaf reflectance spectrum as a function of different parameters: chlorophyll concentration Cab, leaf water content Cw, dry matter concentration Cm, and a leaf structure parameter N. In order to study the influence of the soil reflectance spectrum a soil wetness index was used to shift from a dry soil profile to a wet soil profile (weighted average). Each parameter was varied along its full range of possible variation (e.g. planophile, erectophile, uniform, spherical, plagiophile, etc. for the LIDF, or 10 to 90 mm/cm2 for the chlorophyll concentration). In general, the ratio appears to be far less sensitive to input parameters than the simulated albedo taken alone. The correction ratio appears to be sensitive to LAI, fraction of diffuse radiation and leaf water content only. The ratio is insensitive to the soil moisture, soil albedo, solar inclination, leaf angle distribution, chlorophyll concentration, and leaf structure parameter. This sensitivity to the fraction of diffuse radiation justified the determination of diffuse and direct solar incident spectra. Skye albedos are systematically greater than Kipp ones on vegetation canopies because the average leaf reflectance is lower in the second part of the spectrum (typically 0.15 for 1100–2500 nm, and 0.30 for 320–1100 nm). This explains why in general the total shortwave albedo (typically 0.20 for 320–2500 nm) is lower than the Skye albedo (around 0.30 for 400–1100 nm). This also explains that parameters which have an influence only on the common part of the spectrum of both instruments will have no influence on the resulting ratio (this is the case for LIDF, chlorophyll, and soil reflectance, for instance). On the other hand, leaf water content influences the leaf reflectance spectrum beyond 1100 nm and is a sensitive parameter. The correction is highly sensitive to the LAI: the higher the LAI, the higher the correction, because the canopy spectra look more like a leaf spectrum (with lower reflectance in the 1100–2500 nm region than in the 320–1100 nm region, as seen above). The incoming radiation is also an important parameter for the correction because there is a distinct pattern between direct and diffuse radiation spectra after 1100 nm (see Fig. 5).

615

the correction factor was proved to be insensitive to the Leaf Inclination Distribution Function (LIDF), chlorophyll concentration (except for very yellow leaves) and leaf structure parameter, which are all difficult parameters to obtain). – Soil reflectance spectra: two soil reflectance spectra were measured on an Alpilles soil, in wet conditions and in dry conditions. The real soil reflectance is calculated as a weighted combination of the wet and dry spectra using the soil water content (measured using TDR probes). The sensitivity study, however, proved that the correction factor is totally insensitive to soil moisture (and soil albedo) when the LAI is greater than 1, and largely insensitive to soil moisture even for lower LAI. – Leaf area index (LAI): the LAI was measured on all fields with a LI-COR LAI-2000 device.

3. RESULTS Figure 7 presents the albedos on field 120 (Skye albedos), before and after correction, compared with albedos measured on field 101 (Kipp albedos), on 10 selected days. One may note the variation of albedo with sun zenith angle. After correction, for the selected days, both albedos are very close to each other. Corrected albedos for the whole year and comparison with albedos measured on field 120 are presented in Figure 8. Albedos are averaged for each day. Both albedos are in agreement, although not exactly equal, which is not surprising since we are observing two different fields. After correction, the maximum difference between albedos was reduced from 0.08 to 0.03, which is in agreement with the observations and simulations. This result led us to be confident about the correction scheme. The correction scheme was then applied to all fields where Skye radiometers had been used to measure albedos. The effect of the correction is shown in Figure 9 for fields 102, 120 and 214, plotted together with the Leaf Area Index (LAI). It appears clearly that the correction is correlated with the Leaf Area Index (the LAI appeared to be the most sensitive parameter in the sensitivity study): the higher the LAI, the higher the error (which may reach 0.05). For bare soils, the correction is much smaller.

2.5 Soil and vegetation characteristics The inputs required by the SAIL model are detailed below: – Leaf reflectance spectra: a wheat reflectance spectrum was measured on a test field. No spectrum is available for sunflower fields. We used in a first approach the same wheat profile, but a sunflower-type leaf reflectance spectrum obtained with a leaf radiative transfer model (or a measured one when available) could be used to perform more accurate corrections. As the sensitivity study demonstrates, however, as far as leaves are concerned, only the leaf water content is important for the correction factor (fortunately,

4. SUMMARY AND CONCLUSIONS Silicon sensor radiometers (Skye radiometers) were used to measure crop albedos during the ReSeDA experiment (1996-1997). These radiometers, originally designed to measure sky radiation, are not well adapted to measuring crop canopy albedos. A correction scheme was developed to transform the Skye albedos into true albedos. Since no coincident measurements were available (Skye vs. Kipp), no empirical relationship was used. Furthermore, such a relationship would have been rather specific. Instead, a radiative transfer

616

C. François et al.

0.4 Before correction (zoom)

Albedo

0.3 0.2 0.1 Measurements field 101 (Kipp) Measurements field 120 (Skye)

0 10

11

12

13

14

15

16

17

18

19

20

April 1997 0.4 After correction (zoom)

Albedo

0.3 0.2 0.1

Measurements field 101 (Kipp) Corrected measurements field 120 (transformed into Kipp)

0 10

11

12

13

14

15

16

17

18

19

20

April 1997

Figure 7. Albedo variations before (top) and after (bottom) correction. Albedo is minimum at noon and increases in the morning and afternoon. Kipp and Skye measurements are compared for two nearby wheat fields. After correction, both albedos agree.

0.4 0.35

Before correction

Albedo

0.3 0.25 0.2 0.15 0.1

Measurements field 101 (Kipp) Measurements field 120 (Skye)

0.05 0

Jan

Feb

Apr

Mar

May

Jun

Jul

1997 0.4

After correction

0.35

Albedo

0.3 0.25 0.2 0.15 0.1

Measurements field 101 Kipp) Corrected measurements field 120 (transformed into Kipp)

0.05 0

Jan

Feb

Mar

Apr

May

Jun

Jul

1997

Fig. 8. Same as Figure 4, but zoomed out. Albedos are averaged per day. Albedos after correction agree well between field 101 and field 120. The maximum difference decreases from 0.08 to 0.03 and is now in agreement with the LAI difference between the two fields.

617

Conversion of 400–1100 nm vegetation albedo

⌬albedo

0.1

Field 102

0.05 0

Albedo correction LAI*0.01

-0.05 -0.1

Nov

Dec

Jan

Feb

Mar

Apr

May

Jun

Jul

Aug

Sep

Oct

Nov

Dec

Jun

Jul

Aug

Sep

Oct

Nov

Dec

Jun

Jul

Aug

Sep

Oct

Nov

Dec

1997

⌬albedo

0.1

Field 120

0.05 0 -0.05 -0.1

Nov

Dec

Jan

Feb

Mar

Apr

May

1997

0.1

⌬albedo

Field 214 0.05 0 -0.05 -0.1

Nov

Dec

Jan

Feb

Mar

Apr

May

1997

Fig. 9. Effect of the correction on three fields where Skye radiometers were used (including field 120). Dalbedo = Corrected albedo – Skye albedo. The correlation between the leaf area index (LAI*0.01) and the correction appears clearly. For bare soils, the effect of using Skye radiometers is small. In general, the albedo correction ranges between 0 and 0.05.

model into the canopy (SAIL) was used. It was coupled to a leaf radiative transfer model (PROSPECT) to perform a preliminary sensitivity study. To correct the Skye albedos, a radiative transfer model into the atmosphere (6S) was also used to obtain precise inputs of diffuse and direct radiation spectra (due to the importance of the fraction of diffuse radiation shown by the sensitivity study). Averages of direct and diffuse atmospheric spectra were obtained regardless of atmospheric and illumination conditions. Skye albedos are systematically greater than Kipp ones on vegetation canopies for one principal reason: the average leaf reflectance is lower after 1100 nm. The sensitivity study demonstrated that, among the input parameters, only the LAI, fraction of incident diffuse radiation and leaf water content had an influence on the correction factor, because they had a non-negligible impact on the canopy reflectance after 1100 nm. The application of the correction scheme appeared to correct for the differences observed between fields 101 and 120. If coincident measurements using Skye and Kipp radiometers were available, it would be possible to perform a real validation of the correction scheme. The results obtained on fields 101 and 120, however, already show the efficiency of the method: the maximum difference decreases from 0.08 to

0.03, which is much more acceptable given the small difference in LAI between the two fields. The correction scheme was therefore applied to all fields where Skye radiometers had been used. The correction ranged between 0 (bare soil) and 0.05 (high LAIs). The corrected albedos were successfully compared with albedos derived from POLDER airborne measurements [2], and no difference in the estimation quality was observed when compared with albedos measured on fields where Kipp devices were used. The correction scheme, however, requires some external measurements (LAI, for instance) and is limited to favorable conditions (clear days, homogeneous crop canopies). The application of the correction scheme to other canopies such as forests or very heterogeneous canopies might be questionable. Therefore, the use of Skye radiometers for albedo measurements, despite their lower cost, cannot be recommended. This study demonstrates the usefulness of radiative transfer models for solving an instrumental problem. It also allowed the computation of “universal” diffuse and direct solar radiation spectra that enter most radiative transfer schemes. This study showed that the model outputs (albedos or reflectances) are sensitive to these inputs.

618

C. François et al.

REFERENCES [1] Chédin A., Scott N.A., Wahiche C., Moulinier P., The improved initialization inversion method: A high resolution physical method for temperature retrievals from TIROS N series, J. Climate Appl. Meteorol. 24 (1985) 128–143. [2] Jacob F., Olioso A., Weiss M., Baret F., Hautecœur O., Mapping short-wave albedo of agricultural surfaces using airborne PolDER data, Remote Sens. Environ. 80 (2002) 36–46. [3] Jacquemoud S., Baret F., PROSPECT: a model of leaf optical properties spectra, Remote Sens. Environ. 34 (1990) 75–91. [4] Liang S., Strahler A.H., Walthall C., Retrieval of land surface alebdo from satellite observations: a simulation study, J. Appl. Meteorol. 38 (1999) 712–725. [5] Olioso A., Prevot L., Baret F., Chanzy A., et al., Spatial aspects in the Alpilles-ReSeDA project, in: Marceau D. (Ed.), Scaling and Modelling in Forestry: Application in Remote Sensing and GIS, Université de Montréal (Québec), 1998, pp. 92–102.

[6] Prévot L., Baret F., Chanzy A., Olioso A., et al., Assimilation of multi-sensor and multi-temporal remote sensing data to monitor vegetation and soil: the Alpilles-ReSeDA project, in IGARSS’98 (Seattle, WA, USA), IEEE, Ed., Vol. CDRom, 1998. [7] Sellers P.J., Remote Sensing of the land surface for studies of global change (International Satellite Land Surface Climatology Project Report), Columbia, MD: NASA/GSFC. [8] Spitters C.J.T., Toussaint H.A.J.M., Goudriaan J., Separating the diffuse and direct component of global radiation and its implications for modeling canopy photosynthesis. Part I. Components of incoming radiation, Agric. Forest Meteorol. 38 (1986) 217–229. [9] Verhoef W., Light scattering by leaf layers with application to canopy reflectance modeling: the SAIL model, Remote Sens. Environ. 16 (1984) 125–141. [10] Vermote E., Tanré D., Deuze J.L., Morcrette J.J., Second simulation of satellite signal in the solar spectrum: an overview, IEEE Trans. Geosci. Remote Sens. 35 (1997) 675–686.

To access this journal online: www.edpsciences.org