int. j. remote sensing, 1999, vol. 20 , no. 5, 879± 894

Seasonal variation of leaf chlorophyll content of a temperate forest. Inversion of the PROSPECT model V. DEMAREZ, J. P. GASTELLU-ETCHEGORRY, E. MOUGIN, G. MARTY, C. PROISY Centre d’Etude Spatiale de la BIOspheÁre (UPS/ CNRS/ CNES), 18 Avenue Edouard, Belin BP 31055, Toulouse, France; e-mail: [email protected]

à NE and V. LE DANTEC E. DUFRE Laboratoire d’Ecophysiologie Ve ge tale (CNRS URA2154), Universite de Paris XI, 91405 Orsay Cedex, France

(Received 16 September 1997; in ® nal form 23 June 1998 ) This paper presents part of a 7-month ® eld and laboratory experiment over the deciduous forest of Fontainebleau. Leaf visible and near infrared optical properties of three tree species (oak, beech and hornbeam) were measured each month between April and October 1996. We distinguished the cases of sun and shade leaves, and also abaxial and adaxial leaf surfaces. Spectra were analyzed with reference to leaf chlorophyll content and leaf mass per area. As expected, we observed strong variations of leaf optical properties during the season, with diVerences between sun and shade leaves and abaxial and adaxial surfaces. We also investigated how leaf re¯ ectance and transmittance can provide realistic information about the seasonal variation of leaf chlorophyll content. For that, we used a leaf optical properties model: the PROSPECT model. Inversion of this model with leaf spectra led to the seasonal variation of leaf chlorophyll concentration (m g cmÕ 2 ) we compared with ground measurements. The analysis of spectral data showed that leaf chlorophyll concentration increases strongly at the beginning of the growing season (from April to May), remains stable during several months (from June to August), and decreases strongly when leaves are senescent (from September to October/ November). Chlorophyll concentration of sun leaves tends to be always larger than that of shade leaves. Moreover, chlorophyll concentration depends on leaf species, with oak leaves having the largest chlorophyll concentrations. Abstract.

1.

Introduction

Biochemical components (chlorophyll, nitrogen, etc.) of forest canopies are among essential parameters that control physiological processes. A global and seasonal survey of these parameters is required for understanding forest functioning. In this context, remote sensing can play a unique and essential role because of its capacity to acquire synoptic information at diV erent time and space scales. The concept of spectrometric remote sensing is built on the occurrence of spectral absorption features due to biochemical contents that we want to estimate. It emerges that foliar International Journal of Remote Sensing

ISSN 0143-1161 print/ ISSN 1366-5901 online Ñ 1999 Taylor & Francis Ltd http:// www.tandf.co.uk/ JNLS/ res.htm http:// www.taylorandfrancis.com/ JNLS/ res.htm

880

V. Demarez et al.

re¯ ectance and transmittance are partly related to the concentration of biochemical contents (Brakke et al. 1993, Fourty et al. 1996, Gastellu-Etchegorry et al. 1995, Hosgood et al. 1995, Peterson et al. 1988). For example, most visible radiation (400± 700 nm) incident on fresh leaves is absorbed by leaf pigments. These are essentially chlorophyll a and b with absorption peaks in the blue and red spectral regions, which explains the usual green colour of leaves. Vogelmann and BjoÈrn (1986 ) showed that absorption occured mainly in palisade cells where chloroplasts are numerous. Other pigments such as carotenoids absorb also in the blue. Their in¯ uence becomes visible during leaf senescence due to breakdown of chlorophylls (Sanger 1971). At this stage, leaves are yellow. When the plant dies, brown pigments appear: the leaf re¯ ectance and transmittance decrease regularly between 750 and 400 nm (Trigui and Baldy 1983). In the near-infrared domain (700± 1300 nm), pigments do not absorb and light is scattered by refractive index discontinuities, (1) at the interface of hydrated cell walls with intercellular air spaces, and (2) among cellular constituents (membranes versus cytoplasm) (Gausman 1974, Gausman et al. 1974). Abundance of intercellular spaces tends to increase leaf re¯ ectance and decrease leaf transmittance (Gausman et al. 1970, Boyer et al. 1988). Light scattering depends also on leaf mesophyll arrangements (palisade versus spongy tissue), leaf maturation, and physiological stresses (water, salinity, etc.) that modify biochemical, physiological and structural components (Gausman et al. 1970, Belanger 1990). In order to assess the potential of spectrometric and microwave remote sensing for studying the dynamic and functioning of a temperate deciduous forest, the European Space Agency (ESA) organized the EMAC (European Multisensors Airborne Campaign) over the forest of Fontainebleau in 1994. The forest is located near Paris and is managed by the French O ce for Forestry. In this work, we focus on the seasonal trends of visible and near-infrared optical properties of leaves and their relationships with leaf biochemical (chlorophyll ) and morphological characteristics (thickness and LMA: leaf mass per area ). Re¯ ectance and transmittance measurements were made with a spectrometer. We report here part of the 7-month ® eld and laboratory experiment that took place from April to October 1996, following the EMAC. Three tree species were systematically sampled: oak, beech and hornbeam. We distinguished sun and shade leaves because their optical properties are assumed to diV er markedly (Gausman 1984 ). After a brief analysis of leaf re¯ ectance and transmittance spectra, we present a modelling approach for determining leaf chlorophyll concentration from leaf spectrometric measurements. It relies on the inversion of the PROSPECT model (Jacquemoud and Baret 1990). 2.

M aterials and method

2.1. Study area Our study area is located within the Fontainebleau forest. This is a large deciduous forest (17 000 ha), mainly composed of oaks (Quercus petraea Liebl. and Q. robur L.), beeches (Fagus sylvatica L.), Scots pine (Pinus sylvestris L.), sometimes mixed with birches (Betula sp.), and hornbeams (Carpinus betulus L.). This site was selected within the frame of the French International Geosphere and Biosphere Program because it captures major characteristics of a temperate deciduous forest. On the whole, standing biomass ranges from 0 to 420 tons of dry

881

V ariation of leaf chlorophyll content of a temperate forest

material/ hectare and Leaf Area Index (LAI) ranges from 1 to 8 for the deciduous trees (Le Dantec 1995). 2.2. Field measurements Many measurements were performed during the EMAC (1994): 3000 samples ( leaf, branches, etc.) within 51 forest stands were collected for assessing leaf characteristics such as LMA and chlorophyll concentration. In 1996, measurements were carried out each month between April and October, to assess the seasonal variability of leaf chlorophyll concentration, LMA and leaf thickness. For that, two test sites were selected: an oak (C08 ) and a beech (H13 ) stand. C08 is a mature forest composed of oaks at the dominant story and hornbeams at the dominated story. H13 is a dense and homogeneous pole stand of beeches. Our approach comprised two major steps: collection of leaves in the ® eld, and spectrometric measurements on these leaves in laboratory. Sampling dates were 23 April, 10 May, 17 June, 18 July, 22 August, 30 September and 30 October. Table 1 shows the main structural characteristics of stands C08 and H13 . We randomly selected ® ve trees per species in each stand. For each species (oak, beech and hornbeam) we collected, with the help of a gun, 10 leaves at two canopy levels (top and bottom). At the top, we selected outer leaves on long shoots which received a maximum irradiance and can be considered as sun leaves. At the bottom, all leaves were shaded according to the continuity of the canopy and the high LAI measured in both stands (6.7 for C08 and 7.6 for H13 ). Only one ¯ ush was observed in both stands for each species and consequently all leaves of one species are even-aged. Each month, 60 leaves were collected for further spectrometric measurements. This sample is called hereafter `sample A’. As soon as leaves were collected they were kept in plastic bags at low temperature (# 5ß ), before being measured in the laboratory 24 h later. Measurements of total chlorophyll concentration were conducted with a Minolta SPAD-502 leaf chlorophyll meter (Markwell et al. 1995 ), calibrated against organic extraction analysis (80% acetone, Porra et al. 1989). Empirical relationships (1) between total leaf chlorophyll concentration and SPAD measurements (S ) for both oaks (mature trees) and beeches (both mature trees and saplings) have the same mathematical expression: Chlorophyll (mg cmÕ

2

)= a Ö S / (bÐ

S)

(1)

Parameters a and b are ® tted, respectively, for oaks, beeches and total data set (table 2). Despite diV erences observed between parameters a and b for each species, the predicted chlorophyll content with speci® c parameter is similar to the one predicted Table 1. Age (year), dry matter (DM) by hectare, dominant height (Hd), basal area (G), density of stems by hectare (D), diameter at breast height (DBh) and maximum values of LAI (LAImax ) for oaks (C8 ) and beeches (H13 ) stands. Age

DM (thaÕ 1 )

Hd (m)

G (m2 haÕ

C08

200

409

39

33

521

197

H13

35

127

14

22

4954

60

1

D ) (stemhaÕ

1

)

DBh (mm)

LAImax 4 (oaks) 2.7 (hornbeams) 7.6

882

V. Demarez et al.

Table 2. Coe cients (a, b ) of chlorophyll predictive equations (1) for oak and beech samples. 2 r correlation coe cient. Species Oak Beech Oak+ beech

a

b

r

2

Number of leaves sampled

70.97 62.81 63.92

89.47 81.68 82.85

0.96 0.94 0.94

20 50 70

by the `general’ relationship (diV erences are less than 10% within the range of 2 chlorophyll concentration considered in this study: 10± 90 mg cmÕ ). Consequently the `general’ relationship (oak+beech) is assumed to provide a correct estimation of leaf chlorophyll concentration for the diV erent species studied here. Measurements of chlorophyll concentration on sample A could not be made for other dates than August. Measurements of chlorophyll concentration, LMA and thickness were realized each month, on another set of leaves called `sample B’. Five trees per species were sampled: 20 leaves were collected (10 sun leaves and 10 shade leaves) on each tree. For each leaf we made successively: (i)

six chlorophyll measurements regularly distributed along the lamina using SPAD; (ii) one leaf area measurement using an area-meter (Delta-T devices); (iii) at least two thickness measurements in the middle of the lamina between two secondary veins, using an electronic calliper (resolution 0.01 mm); (iv) ® nally the leaf was dried (oven, 65ß C) and weighed. In April hornbeam leaves were not collected and as budburdst was just occurring we simply mixed all top and bottom leaves of each species ( beech and oak). This explains why hereafter the optical properties of sun and shade leaves in April are identical both for oaks and beeches. Moreover, hornbeam trees in C08 stand are only present in the dominated layer and consequently no true sun leaves are present on these trees: analysis of hornbeam spectra of top and low leaves revealed similar optical properties. Thus no distinction was made between hornbeam leaves at diV erent levels. 2.3. L aboratory optical measurements First, upper (adaxial ) and lower (abaxial ) surface transmittance and re¯ ectance were measured on each leaf of `sample A’. Then we observed that leaf transmittance did not depend on the leaf side which was illuminated (data not shown) so transmittance was measured while irradiating the abaxial surface only. In a ® rst approximation, we assumed that leaf re¯ ectance and transmittance could be described as the sum of two components: a diV use component, supposed to be lambertian, and a specular component, supposed to be mostly directional around the so-called specular direction (Brakke 1994 ). Only diV use re¯ ectance and transmittance result from interactions with mesophyll and thus are in¯ uenced by chlorophyll concentration. Specular re¯ ectance is related to leaf surface interaction only. Our objective being to study leaf chlorophyll content, we devised an experimental protocol intended to measure diV use re¯ ectance and transmittance only. This relies on a single directional measurement out of the angular cone where specular re¯ ection takes place.

V ariation of leaf chlorophyll content of a temperate forest

883

Leaf transmittance and re¯ ectance were measured with an Eotech² spectrometer CCD 512. This instrument covers a spectral region between 400 and 900 nm. Its spectral resolution is 1.1 nm. Leaves were placed in an horizontal sample holder in the centre of a goniometer. Thus, the spectrometer viewed either the leaf or a BaSO4 standard re¯ ectance surface. In this con® guration, the area of the observed surface 2 on the leaf was 2 cm . The radiation source was a 150 W halogen collimated (Ô 7ß ) lamp. We used two diV erent illumination con® gurations for measuring leaf re¯ ectance and transmittance. We measured leaf nadir re¯ ectance with the leaf being irradiated with a 45ß incidence angle. This con® guration allowed us to avoid the in¯ uence of leaf specular re¯ ectance. An isotropic irradiation scheme was employed for measuring transmittance. This was made from a BaSO4 lambertian surface, illuminated with a 45ß incidence angle, which irradiated leaf abaxial surfaces. We determined the relative accuracy of the instrumental con® guration for measuring leaf re¯ ectance and transmittance with the help of repetitive measurements. The precision on measurements is about 2% in the visible and about 0.3% in the near-infrared. 2.4. Modelling approach Chlorophyll concentrations measured in the ® eld with the SPAD were compared with those estimated with the PROSPECT model (Jacquemoud and Baret 1990) used in a backward mode. This model simulates upward and downward hemispherical ¯ uxes that propagate within the leaf and exit the leaf. This is done assuming that the leaf is a stack of N identical elementary layers separated by N ± 1 air spaces. This concept was introduced by Allen et al. (1973) and Gaussman et al. (1970) with the Void Area Index (VAI): VAI= N ± 1, which is an extension of the WillstaÈtter± Stoll theory (WillstaÈtter and Stoll 1918). The number of layers mimics the scattering processes within the leaf internal structure. It is independent of wavelength. Layers are de® ned by their refractive index n (l) and an absorption coe cient K (l), which is assumed to be a linear combination of the speci® c absorption coe cients K j (l) of each absorbing material j , weighted by its concentration: K (l)= ž

j

J

=1

K j (l) C j

(2)

The dimension of concentration is micrograms per unit area. Actually, the major absorbing element in the 400± 900 nm spectral domain is chlorophyll. Thus, we have K (l)# K ab (l) C ab

(3)

where the index ab is relative to chlorophylls a and b . With the assumption that during the growing season parameters n (l) and K ab (l) do not depend on time and on leaf species, leaf re¯ ectance and transmittance are only dependent on C ab and N . Our objective was to determine C ab by inverting the PROSPECT model from our spectral measurements. For that, we used the n (l) and K ab (l) parameters provided by the PROSPECT model. Model inversion consists in determining the input variables of the model that minimize the distance between the measured and the simulated leaf optical properties (Baret 1994). Let us consider a model M that relates the vector X of input variables to the vector Y of output variables: Y = M ( X , Z ). Z represents complementary input ² Eotech is an E.T.A Optic spectrometer (Germany).

884

V. Demarez et al.

variables. Here, output variables are the leaf hemispherical re¯ ectance R (l) and transmittance T (l), whereas input variables X are N and K (l). The inversion procedure relies on the minimization of a merit function (e) that is a simple quadratic sum of the residual between measured and simulated spectra: e= ž

I i

=

1

2 [Y i Õ M (X i , Z i )]

(4)

where I is the number of spectral measurements, Y i is the measured variable, i.e. R (li ) and/ or T (li ), and M (X i, Z i ) is the variable modelled with the set of variables (X i , Z i ). The minimization procedure usually requires an iterative approach. This is ® rst initiated with an approximate assessment of the parameter set X . Here, we used the simplex algorithm (Press et al. 1992). We devised a two-step procedure, in order to reduce computer time and to obtain more robust results (Jacquemoud 1992, Fourty et al. 1996 ). In a ® rst step we inverted the PROSPECT model within the 800± 900 nm spectral domain in order to compute the N parameter alone. All available wavelengths were used between 800 and 900 nm. Once N was determined, we used wavelengths in the visible region in order to retrieve chlorophyll concentration. In both cases the inversion was conducted on leaf transmittance only because the latter proved to be less noisy than the re¯ ectance. 2.4.1. Comparison between SPAD and PROSPECT chlorophyll concentration on `sample A’

Measurements of leaf optical properties and leaf chlorophyll concentration could be measured simultaneously on the same leaves in August only. So, at this date, actual comparison could be conducted between leaf chlorophyll concentration estimated with the PROSPECT model and SPAD chlorophyll measurements. Similarly, we compared the structural parameter N derived from the PROSPECT inversion with measurements of leaf thickness. The inversion procedure for determining parameter N and chlorophyll concentration from the inversion of the PROSPECT model was applied to each leaf of the ® ve diV erent sets gathered in August. 2.4.2. Seasonal variation of N and chlorophyll concentration Results concerning the parameter N and chlorophyll concentration obtained in August were judged good enough to conduct PROSPECT inversion for each leaf of sample A. So the inversion of all monthly leaf data sets was conducted with the twostep approach (like in August): once the parameter N was determined for each leaf, the inversion of the PROSPECT model gave the leaf chlorophyll concentration at each date between April and October. Finally, this radiometrically derived seasonal variation of chlorophyll concentration was compared with that measured with the SPAD spectrometer on sample B. Contrary to the comparison realised in August, here the comparison is conducted on independent samples (A and B). 3.

Results and discussion

3.1. Analysis of leaf spectra Spectra of oak leaves measured with the Eotech from April to October are plotted on ® gure 1. They illustrate clearly the large seasonal variations of optical properties due to leaf maturity. Leaf visible and near infrared spectral variations are discussed hereafter in relation with the illumination context, the leaf side (i.e. abaxial

V ariation of leaf chlorophyll content of a temperate forest

885

Figure 1. Mean spectra of oak shade leaves in April, May, July and October (yellow and brown leaf ). (a) Adaxial re¯ ectance. (b) Transmittance.

or adaxial ) and the leaf species. Our objective is to stress the most important trends and to analyse them in relation with leaf characteristics such as chlorophyll concentration and leaf thickness. 3.1.1. V isible (400± 680 nm) Visible spectra of all leaf species show strong and similar variations with time, whatever the leaf side or the leaf illumination context (® gure 2). On average, during the season, the maximum value (i.e. re¯ ectance and transmittance) is more than twice the minimum value. For example, visible re¯ ectance of the adaxial surface of oak leaves is about 4% in July and August and is about 13% in October. Variation with time of visible re¯ ectances and transmittances is characterized by a decrease at the beginning of the growing season, i.e. from April to June, a nearly constant value from June to September and an increase in October. Major diV erences associated with the leaf illumination context, the leaf surface and the leaf species are summarized in ® gure 2. Sun and shade leaves : visible re¯ ectances and transmittances of sun and shade leaves, that belong to the same species and to the same leaf side, show a similar and signi® cant variation with time during the season. Transmittances of sun leaves tend to be smaller than those of shade leaves. The maximum diV erence of transmittance between sun and shade leaves is reached in July/ August both for oak and beech. These diV erences are signi® cant ( p = 0.1). For a given date, spectral diV erences between species for each class of leaves (sun or shade) are generally inferior to the precision of measurements, which means that spectral discrimination of species in the visible seems impossible. Abaxial and adaxial leaf surfaces: abaxial surfaces have systematically larger visible re¯ ectances than adaxial leaf surfaces, whatever the leaf species and the leaf illumination context are. On average, relative diV erences are around 50% for oak, 21% for beech and 35% for hornbeam, which is signi® cant ( p = 0.05). Re¯ ectances of abaxial and adaxial surfaces undergo the same variations with time, with maxima reached at the beginning and end of the season. L eaf morphological characteristics: generally speaking, leaf transmittance decreases when the quantity of intercepting leaf tissues (number of palisade cells, intercellular spaces) and absorbing pigments (essentially chlorophyll and to a lesser extent carotenoõÈ ds) increase. Leaf re¯ ectance increases if the quantity of intercepting

886

V. Demarez et al.

Figure 2. Mean visible (530± 680 nm) transmittance (Tra: ) and re¯ ectance (Ref ) of abaxial ( ) and adaxial ( ) surfaces of oak, beech and hornbeam leaves from April to October.

leaf tissue increases and if the quantity of absorbing leaf tissue decreases. This explains why, at each date, leaves transmittances of each species decrease when LMA and leaf thickness increase (table 3 and ® gure 3). This simply illustrates the fact that denser and thicker leaves tend to have smaller transmittances. Thus, sun leaves have smaller transmittances than shade leaves. Indeed, from May to August, the LMA of sun leaves is systematically larger than that of shade leaves. If we analyse the evolution of leaf re¯ ectance, we can note that LMA and thickness do not seem to be the only in¯ uential morphological parameters in the visible spectral domain. Indeed, leaf re¯ ectance decreases during the growing season,

V ariation of leaf chlorophyll content of a temperate forest

887

Table 3. Mean leaf mass per area (LMA, g mÕ 2 ) and thickness (th, mm) for oak, beech and hornbeam from April to October. Standard deviations are between brackets. May

June

July

August

LMA th

LMA th

Hornbeam

34 0.08 (7) (0.015) 91.6 0.22 (12.7) (0.01) 56.6 0.18 (9.7) (0.02) 75.3 0.13 (22.3) (0.03) 33.6 0.08 (3.9) (0.01)

37.3 0.095 46.1 0.1 40.5 0.1 32.7 0.1 (10.3) (0.01) (7.6) (0.01) (5) (0.02) (6.4) (0.01) 111 0.31 103 0.24 102 0.24 93.3 0.23 (15.3) (0.08) (17.2) (0.02) (14.4) (0.03) (5.8) (0.03) 63.6 0.2 64.7 0.17 64 0.19 61.2 0.16 (14.5) (0.03) (4.7) (0.02) (5.3) (0.03) (6.3) (0.02) 82.2 0.15 85 0.15 88.8 0.11 58.3 0.1 (20.6) (0.01) (16.4) (0.02) (12.5) (0.02) (12) (0.01) 36.3 0.09 42.6 0.1 44.1 0.16 35.5 0.13 (7) (0.02) (11) (0.01) (10.7) (0.03) (8.1) (0.02)

0.1 (0.02) 0.2 (0.04) 0.16 (0.02) 0.13 (0.01) 0.08 (0.01)

LMA th

October

Leaf species LMA th 41 (10) Oak sun 59.5 (5.1) Oak shade 41.8 (3.3) Beech sun 56.7 (8.5) Beech shade 33 (3.3)

LMA th

September

LMA th

Figure 3. Seasonal variation of mean visible transmittance of sun ( ) and shade ( ) oak leaves, LMA of sun (Ð u Ð ) and shade (Ð { Ð ) oak leaves and thickness of sun ( ) and shade (Ð ± ) oak leaves from May to October.

while leaves tend to thicken. This means that while the quantity of intercepting leaf tissue increases, the quantity of absorbing leaf pigment, i.e. chlorophyll, increases even more. Thus, the proportion of absorbing material versus intercepting material increases during the growing season. The leading role of leaf chlorophyll concentration in the visible domain is further stressed by the fact that in October when leaves senesce and chlorophyll becomes degraded, both leaf re¯ ectance and leaf transmittance increase. This variation cannot be explained by leaf thickness as it does not decrease signi® cantly. The larger volume density of chloroplasts within palisade tissue (adaxial side), compared with that of spongy tissue (abaxial side), combined with the fact that spongy tissues give rise to more numerous scattering mechanisms, explains that leaf abaxial surfaces have larger re¯ ectances than leaf adaxial surfaces (Gausman and Allen 1973). The importance of the highly scattering structure of spongy cells is

888

V. Demarez et al.

underscored with the case of senescent leaves for which chlorophyll begins to break down: abaxial surfaces have a markedly larger re¯ ectance than adaxial surfaces. 3.1.2. Near inf rared (780± 880 nm) Similarly to visible spectra, near-infrared spectra of all leaf species show a strong and similar variation with time, whatever the leaf side and whatever the leaf illumination context (® gure 4) are. On average, during the season, the maximum value (i.e. re¯ ectance and transmittance) is around 35% larger than the minimum value. During the ® rst stage of the growing season leaf transmittance decreases while leaf re¯ ectance

Figure 4. Mean near infrared transmittance (Tra: ) and re¯ ectance (Ref ) of abaxial ( ) and adaxial ( ) surfaces of oak, beech and hornbeam leaves, from April to October.

V ariation of leaf chlorophyll content of a temperate forest

889

increases. Then, these optical quantities stabilize. Finally, they decrease when leaves become senescent. DiV erences associated with the leaf illumination context, the leaf surface and the leaf species are summarized in ® gure 4. Sun and shade leaves: in this spectral domain, re¯ ectances of sun leaves tend to be larger than those of shade leaves, but these diV erences are not signi® cant. Similarly to the visible, near-infrared transmittances of sun leaves are systematically smaller than those of shade leaves. Maximum relative transmittance diV erence is reached in May and June for oak (# 20%) and for beech (# 15%). These diV erences are signi® cant ( p = 0.05). Abaxial and adaxial leaf surfaces: contrary to the visible domain, leaf adaxial and abaxial near-infrared re¯ ectances are never very diV erent, with relative diV erences usually less than 10%. L eaf morphological characteristics: comparison of thickness and LMA of oak, beech and hornbeam leaves with associated leaf near-infrared re¯ ectances and transmittances shows that for any date between April and October larger values of leaf thickness and LMA (table 3) imply smaller leaf transmittances and larger leaf re¯ ectances. The only exception is the re¯ ectance and transmittance decrease in October. We suspect that this may be due to erroneous measurements, probably associated with some deterioration during the transport of leaves. 3.2. Inversion of the PROSPECT model Finally, we investigated the capacity of the PROSPECT leaf optical properties to provide valuable information about seasonal variations of leaf thickness and chlorophyll content. Structural parameter N: table 4 shows mean values of parameter N estimated for August by inverting the PROSPECT model on the transmittance leaf spectra. These results were compared with mean thickness values of oak, beech and hornbeam leaves measured in August. We noted a good correlation (r= 0.98, CI (95%) = [0.96 0.99]) between N and leaf thickness. The parameter N used in the PROSPECT model is not a directly measurable physical quantity: it tends to be large when the number of horizontal layers which constitute the leaf increases. This explains why N increases when the leaf grows. Chlorophyll concentration : ® gure 5 and table 5 show chlorophyll concentrations measured with the SPAD instrument and those computed through the inversion of the PROSPECT model. Correlation is quite good (r= 0.90, CI (95%) [0.83 0.95]), 2 with a root mean square error (rmse) equal to 7.34 mg cmÕ . A detailed analysis revealed that chlorophyll concentration estimated with PROSPECT for sun leaves is larger than for shade leaves similarly to SPAD Table 4. Leaf thickness and parameter N , derived from the inversion of the PROSPECT model, of oak, beech and hornbeam leaves, for the month of August. Standard deviations are in brackets. August 1996 Oak sun leaves Oak shade leaves Beech sun leaves Beech shade leaves Hornbeam

Thickness (mm) 0.24 0.17 0.15 0.10 0.10

(0.02) (0.02) (0.02) (0.01) (0.01)

Parameter N (PROSPECT) 1.92 1.61 1.70 1.43 1.37

(0.05) (0.11) (0.12) (0.10) (0.09)

890

V. Demarez et al.

Figure 5. Chlorophyll concentrations measured with the SPAD instrument versus chlorophyll concentrations estimated with the PROSPECT model. Table 5. Mean chlorophyll concentration measured with the SPAD and derived from the inversion of the PROSPECT model. Standard deviations are between brackets. Chlorophyll concentration (m g cmÕ August 1996 Oak sun leaves Oak shade leaves Beech sun leaves Beech shade leaves Hornbeam

2

)

SPAD measurement

PROSPECT inversion

71.9 (11) 49 (11) 43.5 (6) 42 (4) 37.4 (9)

74 (7.1) 68 (7.2) 58 (5.5) 48 (6) 36 (3)

measurements. Moreover chlorophyll estimated for oak sun leaves and hornbeam leaves agreed with measurements. However chlorophyll estimation of oak shade leaves and beech leaves (sun and shade) is systematically larger than measurements. 3.2.1. Seasonal variation of leaf parameter N and chlorophyll concentration L eaf parameter N: monthly estimates of parameter N and its standard deviation are shown on ® gure 6. Values of N ® t pretty well within the usual range ([1.5 2.5]) for dicotyledons (Jacquemoud et al. 1996). It appears that parameter N increases between April and May when leaf tissue gets structured, does not vary so much till September while leaf structure remains stable, and increases during the month of October. This increase is simply linked to the relatively strong decrease of leaf near infrared transmittance in October (® gure 4). This variation could be explained by the fact that during this period leaves senesce, which implies that multiple scattering increases due to the diminution of leaf water content (Jacquemoud et al. 1996 ). This explanation disagrees with the unexplained fact that at the same period we found that leaf near infrared re¯ ectance decreases. Sun leaves have always a larger parameter N than shade leaves, because nearinfrared transmittances of sun leaves are systematically smaller than those of shade leaves. This agrees with the fact that leaves of hornbeam trees, which make up the understorey of C08 stand, have the lower parameter N . Moreover, we can note that parameter N is larger for oak than for beech.

V ariation of leaf chlorophyll content of a temperate forest

891

Figure 6. PROSPECT-derived leaf parameter N of hornbeam leaves (Ð *Ð ), oak sun leaves ( ), oak shade leaves ( ), beech sun leaves ( ) and beech shade leaves ( ), between April and October. Standard deviations are also plotted. Chlorophyll concentration : PROSPECT-derived chlorophyll concentrations are shown in ® gure 7. It appears that the variation of chlorophyll concentration with time is quite classical: strong increase during the ® rst phase of the growing season, followed by a stability between June and September, and a strong decrease in October. Two major results must be stressed:

(1) Sun leaves tend to always have larger chlorophyll concentration than shade leaves. The only exception occurs in May for oak leaves. At this date, oak

Figure 7. PROSPECT-derived leaf chlorophyll concentration (m g cmÕ 2 ) of hornbeam leaves ( ), oak sun leaves (Ð Ð ), oak shade leaves ( ), beech sun leaves ( ) and beech shade leaves ( ), between April and October. Standard deviations are also plotted.

892

V. Demarez et al.

sun and shade leaves have very similar chlorophyll concentration anyway. Actually, this inversion of trend could be very well explained by the variability of chlorophyll measurements (table 5). (2) Chlorophyll concentration depends on leaf species. On average, oak leaves have larger chlorophyll concentrations than beech leaves. For example, 2 between June and September mean values are 70 mg cmÕ for oak and 2 52 mg cmÕ for beech. During that period leaf chlorophyll concentration of 2 hornbeam is much lower, around 38 mg cmÕ . The comparison between the seasonal variation of the chlorophyll content estimated with PROSPECT and that measured with the SPAD spectrometer (® gure 8), reveals that we have the same trends and the same dynamic of values. However we can note that the predicted PROSPECT chlorophyll values are higher than those measured for beech (sun and shade) and shade oak leaves, whereas for sun oak and hornbeam leaves, estimated values are comparable to measurements. 4.

Concluding remarks

This work allowed us to study the seasonal variation of chlorophyll concentration and optical properties of oak, beech and hornbeam leaves of a temperate deciduous forest. Optical properties were assessed with the help of laboratory spectrometric measurements conducted during one year, each month from April to October. We showed that leaf optical properties and chlorophyll concentration depend a lot on leaf species and sun illumination context (sun leaves versus shade leaves). Moreover, re¯ ectance depends also, to a lesser extent, on the leaf side (adaxial surface versus abaxial surface). We compared the leaf chlorophyll concentrations measured with the SPAD spectrometer and the leaf chlorophyll concentrations predicted with the PROSPECT model. SPAD measurements and PROSPECT estimates were realized on the same

Figure 8. Measured leaf chlorophyll concentration (m g cmÕ 2 ) of hornbeam leaves ( ), oak sun leaves (Ð Ð ), oak shade leaves ( ), beech sun leaves ( ) and beech shade leaves ( ), between April and September. Standard deviations are also plotted.

V ariation of leaf chlorophyll content of a temperate forest

893

leaves, in August. This work revealed good agreement between measured and predicted leaf chlorophyll concentrations for sun oak and hornbeam leaves. However, for beech (sun and shade) and shade oak leaves, predicted chlorophyll values were higher than the measured ones. The same trends were observed for the seasonal chlorophyll variation. For the seasonal study, SPAD measurements and PROSPECT estimations were realized on independent samples. The fact that similar results were obtained with measurements realized on the same samples (August) and on independent samples (seasonal study) stresses the robustness of the sample method. DiV erences observed between predicted and measured chlorophyll values could be explained by the fact that (1) the spectral measurements may not be signi® cant enough to represent the spectral response of the leaf, and/ or that (2) absorbing coe cients K ab proposed by the PROSPECT model may not be well adapted to our leaf species. Concerning the ® rst explanation we found two limitations: (1) it is likely that an integrating leaf 2 surface of 2 cm (contrary to SPAD measurements) is too small to take the heterogeneity of the spatial distribution of chlorophyll within the leaf into account, and (2) our directional re¯ ectances and transmittances are not representative enough of hemispherical optical properties. Acknowledgments

We acknowledge the European Space Agency (ESA), the French National O ce of Forestry (ONF), the French Ecosystem Committee `Foreà ts tempe re es’, the French Space Center (CNES), the National Program of Remote Sensing (PNTS), and the Program of Environment (SEAH) for their collaboration and support. We also thank people from LEV for ® eld studies. References A llen, W . A ., G ausman, H. W., and Richardson, A. J., 1973, WillstaÈtter± Stoll theory of leaf re¯ ectance evaluated by ray tracing. Applied Optics 12, 2448± 2453. Baret, F., 1994, Use of re¯ ectance spectral variation to retrieve canopy biophysical characteristics. In Advances in Environmental Remote Sensing, edited by F. M. Danson and S. E.

Plummer (Chichester: Wiley), pp. 35± 51. Belanger, M. J., 1990, A seasonal perspective of several leaf developmental characteristics as related to the red edge of plant re¯ ectance, Ph.D. Thesis, York University, Ontario, 110 pp. B oyer, M ., M iller, J ., B elanger, M ., H are, E. and Jiyou, W., 1988, Senescence and spectral re¯ ectance in leaves of Northern Pin Oak (Quercus palustris Muenchh). Remote Sensing of Environment, 25, 71± 87. Brakke, T. W., 1994, Specular and diVuse components of radiation scattered by leaves. Agricultural and Forest Meteorology , 71, 283± 295. B rakke, T . W ., W ergin, P . W ., E rbe, E. F., and Harnden, J. M., 1993, Seasonal variation in the structure and red re¯ ectance of leaves from Yellow Poplar, Red Oak, and Red Maple. Remote Sensing of Environment, 43, 115± 130. F ourty, T ., B aret, F ., J acquemoud, S ., S chmuck, G., and V erdebout, J., 1996, Leaf optical properties with explicite description of its biochemical composition: direct and inverse problems. Remote Sensing of Environment, 56, 104± 117. G astellu-E tchegorry, J . P ., Z agolski, F ., M ougin, E ., M arty, G., and G iordano, G., 1995, An assessment of canopy chemistry with AVIRISÐ a case study in the Landes forest, France. International Journal of Remote Sensing, 16, 487± 501. G ausman, H. W., 1974, Leaf re¯ ectance of near-infrared. Photogramm etric Engineering and Remote Sensing, 40, 183± 191. G ausman, W., 1984, Evaluation of factors causing re¯ ectance diVerences between sun and shade leaves. Remote Sensing of Environment, 15, 177± 181.

894

V ariation of leaf chlorophyll content of a temperate forest

G ausman, H. W., and A llen, W. A., 1973, Optical parameters of leaves of 30 plant species. Plant Physiology , 72, 57± 62. G ausman, H . W ., A llen, W . A ., C ardenas, R., and R ichardson, A. J., 1970, Relation of light re¯ ectance to histological and physical evaluation of cotton leaf maturity. Applied Optics 9, 545± 552. G ausman, H . W ., A llen, W. A., and E scobar, D. E., 1974, Refractive index of plant cell walls. Applied Optics, 13, 109± 111. H osgood, B ., J acquemoud, S ., A ndreoli, G ., V erdebout, J ., P edrini, G., and S chmuck,

G., 1995, Leaf optical properties experiment (LOPEX 93), European Commission, Joint Research Center, Institute For Remote Sensing Applications, Ispra (Italy), Report EUR-16095-EN. J acquemoud, S., 1992, Utilisation de la haute re solution spectrale pour l’e tude des couverts ve ge taux: de veloppement d’un modeÁle de re ¯ ectance spectrale, TheÁse de doctorat de l’Universite Paris VII, 92 pp. J acquemoud, S., and B aret, F., 1990, PROSPECT: a model of leaf optical properties spectra. Remote Sensing of Environment, 34, 75± 91. J acquemoud, S ., U stin, S .L ., V erdebout, G ., S chmuck, G ., A ndreoli, G., and H osgood, B., 1996, Estimating leaf biochemistry using the PROSPECT leaf optical properties model. Remote Sensing of Environment, 56, 194± 202. L e D antec, V., 1995, Utilisation de la te le de tection multispectrale en vue de mode liser le bilan carbone d’un massif forestier, D.E.A d’Ecologie Ge ne rale et Production Ve ge tale, Universite Paris Sud XI, 35 pp. M arkwell, J ., O sterman, J. C., and M itchell, J. L., 1995, Calibration of the Minolta SPAD-502 leaf chlorophyll meter. Photosynthesis Research , 46, 467± 472. P eterson D . L ., A ber J . A ., M atson P . A ., C ard D . H ., S wanberg N ., W essman C., and S panner M., 1988, Remote sensing of forest canopy and leaf biochemical contents. Remote Sensing of Environment, 24, 85± 108. P orra, R . J ., T hompson, W. A., and K riedman, P. E., 1989, Determination of accurate extinction coe cients and simultaneous equations for assaying chlorophylls a and b extracted with four diVerent solvents: veri® cation of the concentration of chlorophyll standards by atomic absorption spectroscopy, Biochemical Biophysical Acta, 975, 384± 394. P ress, W . H ., T eukolsky, S . A ., V etterling, W. T., and F lannery, B. P., 1992, Numerical recipies in C. In T he Art of Scienti® ng Computing , 2nd edn (Cambridge: Cambridge University Press). S anger, J. E., 1971, Quantitative investigation of leaf pigments from their inception in buds through autumn coloration to decomposition in falling leaves. Ecology, 52, 1075± 1089. T rigui, A., and B aldy, C., 1983, Evolution des proprie te s optiques des feuilles d’olivier se nescentes ou malades, 2ieÁme Colloque International de Signatures d’objets en te le de tection, Bordeaux (France), 12-16 September , Les colloques de l’INRA 23:353:363. V ogelmann, T. C., and B jo¨ rn, L. O., 1986, Plants as light traps. Physiological Plant, 68, 704± 708. W illsta¨ tter, R., and S toll, A., 1918, Untersuchunger uÈ ber die Assimilation der KohlensaÈure (Springer, Berlin, 1918).