Remote Sensing of Environment 121 (2012) 415–425

Contents lists available at SciVerse ScienceDirect

Remote Sensing of Environment

journal homepage: www.elsevier.com/locate/rse

Mapping spatio-temporal variation of grassland quantity and quality using MERIS

data and the PROSAIL model
Yali Si a, b,⁎, Martin Schlerf c, 1, Raul Zurita-Milla b, Andrew Skidmore b, Tiejun Wang b
a
Ministry of Education Key Laboratory for Earth System Modeling, and Center for Earth System Science, Tsinghua University, Beijing, 100084, China
b
Faculty of Geo-information Science and Earth Observation (ITC), University of Twente, P.O. Box 6, 7500AA Enschede, The Netherlands
c
Centre de Recherche Public, Gabriel Lippmann, L-4422 Belvaux, Luxembourg

a r t i c l e i n f o a b s t r a c t

Article history: Accurate estimates of the quantity and quality of grasslands, as they vary in space and time and from regional
Received 25 January 2011 to global scales, furthers our understanding of grassland ecosystems. The Medium Resolution Imaging
Received in revised form 10 February 2012 Spectrometer (MERIS) is a promising sensor for measuring and monitoring grasslands due to its high spectral
Accepted 11 February 2012
resolution, medium spatial resolution and a two- to three-day repeat cycle. However, thus far the multi-
Available online 22 March 2012
biome MERIS land products have limited consistency with in-situ measurements of leaf area index (LAI),
Keywords:
while the multi-biome canopy chlorophyll content (CCC) has not been validated yet with in-situ data. This
Grassland study proposes a single-biome approach to estimate grassland LAI (a surrogate of grass quantity) and leaf
Quantity chlorophyll content (LCC) and CCC (surrogates of grass quality) using the inversion of the PROSAIL model
Quality and MERIS reflectance. Both multi-biome and single-biome approaches were validated using two-season
LAI in-situ data sets and the temporal consistency was analyzed using time-series of MERIS data. The single-
Chlorophyll biome approach showed a consistently better performance for estimating LAI (R2 = 0.70, root mean square
MERIS error (RMSE) = 1.02, normalized RMSE (NRMSE) = 16%) and CCC (R 2 = 0.61, RMSE = 0.36, NRMSE = 23%)
PROSAIL
compared with the multi-biome approach (LAI: R 2 = 0.36, RMSE = 1.77, NRMSE = 28%; CCC: R2 = 0.47,
LUT
RMSE = 1.33, NRMSE = 84%). However, both single-biome and multi-biome approaches failed to retrieve
LCC. The multi-biome LAI was overestimated at lower LAI values (b 2) and saturated at higher LAI values
(≥ 4), and the multi-biome CCC was consistently overestimated through the whole data range. Similar
temporal trajectories of grassland LAI and CCC estimates were observed using these two approaches, but
the multi-biome trajectory consistently produced larger values than the single-biome trajectory. The
spatio-temporal variation of grassland LAI and CCC estimated by the single-biome approach was shown to
be closely associated with agricultural practices. Our results underline the potential of mapping grassland
LAI and CCC using the PROSAIL model and MERIS satellite data.
Crown Copyright © 2012 Published by Elsevier Inc. All rights reserved.

1. Introduction
The estimation of the biophysical and biochemical properties of
vegetation has proven useful for a large variety of ecological
applications (Asner, 1998; Houborg et al., 2007). In particular,
accurate estimates of the quantity and quality of grasslands, as they
vary in space and time and from regional to global scales, further
our understanding of grassland ecosystems. Leaf area index (LAI) is
commonly used in ecosystem models because it influences exchanges
⁎ Corresponding author at: Ministry of Education Key Laboratory for Earth System
Modeling, and Center for Earth System Science, Tsinghua University, Beijing, 100084,
China. Tel.: + 86 1 62797419; fax: + 86 1 62797284.
E-mail addresses: yali@itc.nl (Y. Si), schlerf@crpgl.lu (M. Schlerf),
zurita-milla@itc.nl (R. Zurita-Milla), skidmore@itc.nl (A. Skidmore), tiejun@itc.nl
(T. Wang).
1
Previously with ITC, University of Twente, The Netherlands.
of energy, water and carbon between plants and the atmosphere. LAI
quantifies canopy structure, and can be used to predict primary pro-
ductivity and grass growth. Hence LAI can be used as a surrogate of
grass quantity. Changes in leaf chlorophyll content (LCC) provide an
indicator of maximum photosynthetic capacity, leaf developmental
stage, productivity and stress (Horler et al., 1983; Jenkins et al.,
1981; Medina & Lieth, 1963). In addition, chlorophyll content gives
an indirect estimation of the nutrient status because much of leaf
nitrogen is incorporated in chlorophyll (Filella et al., 1995). Spectro-
scopic estimation of nitrogen has always been associated to that of
chlorophyll pigments, based on the fact that the two variables are mod-
erately correlated within and across ecosystems (R2 in the range of
0.4–0.6) (Kokaly et al., 2009; Wright et al., 2004). The canopy chloro-
phyll content (CCC) is defined here by multiplying LAI by LCC. Both
LCC and CCC therefore can be utilized as surrogates of grass quality.
Remote sensing can be used to estimate the spatio-temporal varia-
tion of grassland quantity and quality through repeatable measurement,

0034-4257/$ – see front matter. Crown Copyright © 2012 Published by Elsevier Inc. All rights reserved.
doi:10.1016/j.rse.2012.02.011
416 Y. Si et al. / Remote Sensing of Environment 121 (2012) 415–425
at a relatively low cost compared to physical field measurements
(Mutanga et al., 2004). The Medium Resolution Imaging Spectrometer
(MERIS), operating in the visible and near-infrared reflective spectral
range, has high spectral resolution, moderate spatial resolution and a
two- to three-day repeat cycle. With respect to the spectral resolution,
MERIS provides data in fifteen spectral bands, centered at 412, 442,
490, 510, 560, 620, 665, 681, 709, 754, 760, 779, 865, and 900 nm, with
a typical bandwidth of 10 nm (Rast et al., 1999). These narrow bands
are sensitive to both vegetation structure and biochemistry, which
makes MERIS a promising sensor for the retrieval of vegetation
properties across large geographical scales. MERIS pixel size is approxi-
mately 300 m at full spatial resolution (FR) and 1200 m for reduced
resolution (RR). More precisely, at FR resolution, the pixel size varies
from approximately 260 m to 390 m due to the wide field of view of
the instrument. Finally, the three-day repeat cycle allows tracing vegeta-
tion phenology. In this study we use FR data because using lower spatial
resolution data tends to increase land cover mixture effect and therefore
the complexity of interpreting the results.
Traditional remote sensing methods used for extracting biophysi-
cal and biochemical ecosystem characteristics rely on the observed
spectral features via an empirical relationship linking the variables
of interest to a combination of radiometric measurements (i.e., vege-
tation indices) (Darvishzadeh et al., 2008b; Dorigo et al., 2007; van
der Meer et al., 2001). However, empirical relationships are site,
time, and vegetation specific (Baret & Guyot, 1991; Gobron et al.,
1997). Radiative transfer models aim to generalize empirical results
and improve the reliability of vegetation parameter retrieval, in
order to cope with a wide range of situations. Physically-based
models describe the transfer and interactions of radiation inside the
canopy based on physical laws and offer an explicit connection
between the biophysical and biochemical variables of vegetation
and canopy reflectance (Baret et al., 2000; Darvishzadeh et al.,
2008a; Houborg et al., 2007).
The inversion of the PROSAIL model (Jacquemoud & Baret, 1990;
Kuusk, 1991; Verhoef, 1984, 1985) was selected for this study be-
cause it has been extensively validated using ground, airborne, and
space borne data sets with different sensors/platform such as
MODIS, MERIS, VEGETATION, POLDER, IKONOS, SPOT, TM and
HYPERION (see details in Jacquemoud et al., 2009). Most applications
were focused on crops, forests, or a combination of different vegeta-
tion types and few studies have explored the potential of radiative
transfer models to estimate properties of grassland (Botha et al.,
2010; He & Mui, 2010; Jacquemoud et al., 2009; Maire et al., 2011).
Two studies specifically dealing with grassland successfully retrieved
LAI and CCC at the field level (Darvishzadeh et al., 2008a; Vohland &
Jarmer, 2008). These studies were validated either by a limited
number of samples (Vohland & Jarmer, 2008) or by single-season
in-situ measurements (Darvishzadeh et al., 2008a). As different
plant species show specific morphological and anatomical features,
which again are modified by nutrient supply and phenological
development (Vohland & Jarmer, 2008), it is desirable to extend the
validation procedure by increasing the number of sites and in-situ
measurements over different seasons.
A multi-biome MERIS product has been generated by Bacour et al.
(2006) to estimate global vegetation characteristics, by using the in-
version of the PROSAIL model with a neural network approach. Due
to the limited number of validation points used by Bacour et al.
(2006), it is difficult to draw definitive conclusions about the accuracy
of the multi-biome approach. Previous validation of the multi-biome
LAI against in-situ measurements showed an overestimation at
lower values of LAI and saturation at higher values of LAI (Bacour et
al., 2006; Canisius et al., 2010). It was concluded that none of the
MERIS LAI algorithms currently used meet the performance require-
ments set by the Global Climate Observing System (Canisius et al.,
2010). Moreover, the estimation of the multi-biome CCC has not yet
been validated against in-situ measurements. As the multi-biome
approach operates across different biomes, a single-biome approach is
expected to increase the accuracy in estimating grassland properties.
The objectives of this study were i) to develop a single-biome ap-
proach for the retrieval of grassland properties (i.e., LAI, LCC, and
CCC), using the inversion of the PROSAIL model from MERIS FR imagery;
ii) to compare the performance of the multi-biome and single-biome
approaches against in-situ measurements over two seasons; iii) to
analyze the temporal consistency of the multi-biome and single-
biome approaches using time-series MERIS imagery.
2. Materials and methods
2.1. Study area
The study area is situated in the northern part of The Netherlands,
in the provinces of Groningen and Friesland (Fig. 1). The climate is
temperate and maritime. The average temperature is 2 °C in January
and 19 °C in July, with an annual average of about 10 °C. Clouds gen-
erally appear every day and in the winter fog often abounds. Rainfall
occurs frequently and averages about 765 mm annually. Two types of
grassland are present in the study area. Agricultural grassland, ac-
counting for 70% of the grassland in the field sampling area, is man-
aged by local farmers by regular fertilization, mowing, and cattle
grazing. The remaining semi-natural grassland is managed as natural
reserves and in some of these naturalized cattle is allowed to graze
year round. The plant community in agricultural fields is dominated
by Lolium perenne and Poa pratensis, while in semi-natural areas it is
dominated by Festuca rubra, Puccinellia maritima, Agrostis stolonifera,
Plantago maritima, and Triglochin maritima.
2.2. Sampling design
Fieldwork was conducted in April to June 2008. The mean temper-
ature during the field campaigns of April, May, and June was 8.6 °C,
14.4 °C, and 16.4 °C, and precipitation was 27.7 mm, 6.4 mm, and
51.6 mm. Within the three months, 9 days had a clear sky, 52 days
had less than 50% cloud cover, and 30 days were heavily clouded. A
total of 30 grassland fields were selected, of which 23 are distributed
in agricultural grassland and 7 in semi-natural grassland. Sampling
plots of 300 by 300 m were designed and within each plot, five
quadratic subplots of 1 by 1 m were established, of which one was sit-
uated in the center and four in each corner. A GPS was used to locate
the position of each subplot in the field. Vegetation measurements at
the plot level were obtained by averaging the measurements of the 5
subplots. For each plot, the MERIS pixels intersecting it were averaged
to get a representative MERIS reflectance.
2.3. LAI and chlorophyll measurement
LAI was measured using a Plant Canopy Analyzer LAI-2000 (LICOR
Inc., Lincoln, NE, USA), which determines effective LAI using measure-
ments of diffuse solar radiation above and below the grass canopy.
The LAI was measured under overcast sky conditions between 10:00
and 16:00, using a view restrictor of 90°. The average LAI was
calculated in each subplot, based on one above canopy measurement
and five below-canopy measurements. Special attention was taken to
ensure stable sky conditions between the above- and below-canopy
measurements. When calculating the LAI, none of the outer rings
were eliminated in the gap-fraction inversion. Despite the non-
random distribution of leaves, no corrections for clumping effects
were applied as these were assumed to be compensated by the
overestimation of LAI through grass stems.
A portable SPAD-502 chlorophyll meter (Minolta, Japan) was used
for the measurement of LCC. The SPAD measures a unitless value
which is highly correlated with leaf chlorophyll content. For each sub-
plot, the SPAD reading was calculated based on the average of 30
 Y. Si et al. / Remote Sensing of Environment 121 (2012) 415–425 417

Fig. 1. Location of the study area and the in-situ sampling plots (black dots). The aero-photo, covered by a grid (representing 300 × 300 m MERIS pixels), illustrates the heteroge-
neity level of the landscape at the MERIS level.

randomly selected leaf readings within the subplot. In order to
convert the unitless SPAD readings into LCC (μg cm − 2), a total of 36
samples were collected during the summer campaign from both
type of grasslands (25samples from the agricultural and 11 from the
semi-natural grassland). Samples were measured with SPAD and
subsequently placed in plastic bags inside a portable ice box and
transported to the laboratory for LCC measurements. The total
chlorophyll a and b contents were then extracted using the
dimethyl-sulphoxide solvent method, calculated on the basis of co-
efficients published by Wellburn (1994). In accordance with Mark-
well's formulation (Markwell et al., 1995), an exponential equation
(Cab = 84.8 ∗ exp (0.00702 ∗ SPAD) − 82.01) was found to best de-
scribe the relationship between the calculated LCC (μg cm − 2) and
the SPAD readings (Fig. 2). As the data points were regularly distrib-
uted along the fitted line and no skewed points were observed, we
considered the relationship to be the same for the two grassland
types. The CCC for each subplot was obtained by multiplying LCC
and corresponding LAI. The statistical description of field measure-
ments of LAI, LCC, SPAD, and CCC can be found in Table 1.
2.4. Soil reflectance measurement
The spectral reflectance of bare soil was acquired from vegetation
free sites using a GER 3700 spectroradiometer (Geophysical and Envi-
ronmental Research Corporation, Buffalo, New York). Individual mea-
surements from different sites were then smoothed using a Savitzky–
Golay filter (frame size 15 data points, 2nd degree polynomial)
(Savitzky & Golay, 1964) and resampled to MERIS wavebands. The
mean reflectance of these measurements was calculated to represent
the soil optical properties in the study area.
2.5. MERIS imagery pre-processing
Five MERIS FR L1b top of atmosphere radiance images, captured
on 11th of February, 9th of April, 8th of June, 9th of September, and
26th of December 2008, were acquired respectively. The time span
between image capturing and field data collection is largely influ-
enced by the revisit cycle of the sensor, the weather condition and
the field work logistics. Given the 2–3 day revisit cycle of MERIS and

Table 1
Summary statistics (minimum, mean, maximum, standard deviation, and range) of the
field measurement of grass leaf area index (LAI), leaf chlorophyll content (LCC), SPAD
readings, and canopy chlorophyll content (CCC).

Measured variables Min Mean Max StDev Range

April (n = 30)
LAI (m− 2 m− 2) 0.35 0.9 1.66 0.39 1.32
SPAD (unitless) 18.56 37.13 48.5 6.02 29.94
LCC (μg cm− 2) 21.9 30.16 37.02 3.63 15.12
CCC (g m− 2) 0.1 0.28 0.54 0.14 0.44
June (n = 30)
LAI (m− 2 m− 2) 0.43 3.06 6.77 1.78 6.34
SPAD (unitless) 18.56 37.45 48.67 6.16 30.11
LCC (μg cm− 2) 14.59 28.48 37.34 4.58 22.75
CCC (g m− 2) 0.12 0.86 1.69 0.5 1.57
Pooled (n = 60)
LAI (m− 2 m− 2) 0.35 1.97 6.77 1.67 6.43
SPAD (unitless) 18.56 38.35 48.67 5.44 30.11
LCC (μg cm− 2) 14.59 29.32 37.34 4.18 22.75
Fig. 2. The empirical relationship between SPAD readings and leaf chlorophyll content CCC (g m− 2) 0.1 0.57 1.69 0.47 1.59
based on leaf samples collected during the field campaign in 2008.
418 Y. Si et al. / Remote Sensing of Environment 121 (2012) 415–425

the frequent cloudy weather in the northern part of The Netherlands,
we were able to acquire MERIS images with a maximum of 10 days
difference from the time of field data collection. Since the period is
relatively short and the sudden drop in biomass due to mowing was
excluded by cross-checking with farmers, we assume that the grass-
lands did not change significantly within this period.
The geolocation accuracy of MERIS images is around half a pixel
(170 ± 10 m), meaning that the mismatch error between the sam-
pling plots and MERIS pixels would be around half a pixel. The
MERIS L1b product is geolocated top of atmosphere radiance whereas
the MERIS L2 product is geolocated surface reflectance over sea but
top of aerosol over land. The SCAPE-M (self-contained atmospheric
parameters estimation from MERIS data) algorithm proposed by
Guanter et al. (2008) was utilized to convert level 1b top of atmo-
sphere to top of canopy reflectance, correcting distortions caused by
the interaction between solar radiation and atmospheric components.
The SCAPE-M algorithm derives aerosol optical thickness (AOT), co-
lumnar water vapor (CWV) and reflectance maps from MERIS data
over land in a consistent and automatic manner. The specific proce-
dure includes masking non-land pixels (including cloud pixels), ac-
counting for topographic effects, retrieval of AOT, retrieval of CWV,
and retrieval of surface reflectance. Atmospheric parameters are cal-
culated on a per-pixel basis so that pixel-to-pixel variations in AOT,
CWV, and surface elevation and topography are properly taken into
account.
The MERIS reflectance of each subplot was extracted from the
April and June corrected images. The reflectance at the plot level
was obtained by averaging the reflectance of five subplots. Two ap-
proaches have been proposed for generating MERIS multi-biome
products using either MERIS L1b images or MERIS L2 images as
input (Bacour et al., 2006). L1b images were used to generate the
MERIS multi-biome products in this study, as previous study demon-
strated that LAI estimates were nearly identical (R 2 > 0.98) using ei-
ther L1b or L2 MERIS images as input in the multi-biome approach
(Canisius et al., 2010).
The MERIS bands selected for this study are the same 11 spectral
bands utilized for generating multi-biome products (Bacour et al.,
2006) and are centered at 490, 510, 560, 620, 665, 681, 709, 754,
779, 865, and 885 nm. The remaining 4 bands were removed because
one is an oxygen absorption band (760 nm), one is a water absorption
band (900 nm), and two are strongly affected by atmosphere effects
(412 nm, 442 nm) and only provide marginal information (Clevers
et al., 2007).
2.6. The PROSAIL radiative transfer model
include sun zenith angle, ts (deg); sensor viewing angle, to (deg);
relative azimuth angle, phi (deg); and fraction of diffuse incoming
solar radiation, skyl. A soil brightness parameter, scale, was utilized
to account for the changes induced by moisture and roughness in
soil brightness (Atzberger et al., 2003; Darvishzadeh et al., 2008a).
2.7. The single-biome approach
The single-biome approach is based on a previous study
(Darvishzadeh et al., 2008a) which retrieved biophysical and
biochemical parameters in heterogeneous grassland at the field
level using the inversion of the PROSAIL model. The look-up table
(LUT) inversion method was selected because it yielded good retriev-
al performances in estimating grassland properties at the field level
(Darvishzadeh et al., 2008a; Vohland & Jarmer, 2008). The LUT is a
conceptually simple technique, which potentially overcomes limita-
tions of iterative optimization algorithms (Kimes et al., 2000).
Physically-based modeling can suffer from the ill-posed problem
during the inversion (Atzberger, 2004), where the inversion solution
is not always unique, as various combinations of canopy parameters
may yield almost similar spectra (Weiss & Baret, 1999). Previous
studies suggested that utilizing prior information is an efficient
way of solving this problem and of improving the accuracy of the
estimated canopy variables (Combal et al., 2002). The LUT generated
for the retrieval of grassland properties from MERIS reflectance was
optimized according to prior knowledge and the existing literature
(Bacour et al., 2006; Darvishzadeh et al., 2008a; Vohland & Jarmer,
2008). The geometrical parameters were collected from the metadata
provided with each MERIS image. Thus, the per-pixel sun zenith
angle, sensor viewing angle and relative azimuth angle, driven by
the ENVISAT orbit and MERIS swath, were assigned. The values for
leaf chlorophyll a + b content, dry matter content, equivalent water
thickness, soil brightness parameter, and hot spot size parameter
were defined as for the previous field-level study (Darvishzadeh et
al., 2008a) (Table 2). Considering the changes caused by different
grass species and phenological development, broader ranges of the
leaf angle (20–70°) and LAI (0.1–8.0 m 2 m − 2) were utilized com-
pared to Darvishzadeh et al. (2008a). Phenological information has
been utilized to limit the LAI range to 0.1–4.0 for the cold seasons
Table 2
Ranges for input parameters used for generating look-up table (LUT) using the PROSAIL
model simulating MERIS reflectance in the single-biome approach, in comparison with
a field-level study (Darvishzadeh et al., 2008a).
Parameter Abb. Unit Range or fixed value

A combination of the PROSPECT leaf optical properties model Leaf

Leaf structural parameter N No dimension 1.5–1.9
(Jacquemoud & Baret, 1990) and SAILH canopy reflectance model Leaf chlorophyll content LCC ug cm− 2 15–55
(Kuusk, 1991; Verhoef, 1984, 1985), has been validated for different Equivalent water thickness Cw g cm− 2 0.01–0.02
kinds of vegetation and is therefore considered to be suitable for ap- Dry matter content a Cm g cm− 2 0.0025–0.005
plications in the retrieval of grassland properties in terms of both ac- Canopy
Leaf area index b LAI m2 m− 2 0.1–4.0 (Nov–Apr)
curacy and running time of the inversion process (Jacquemoud et al.,
0.1–8.0 (May–Oct)
2000). This original PROSPECT version was selected to be consistent Mean leaf inclination angle c
ALA Deg 20–70
with a field-level study that demonstrated that grass LAI and CCC Hot spot size parameter hot m m− 1 0.05–0.1
can be successfully retrieved using the inversion of the PROSAIL
model (Darvishzadeh et al., 2008a). Besides, this version of Soil
Soil brightness parameter scale No dimension 0.5–1.5
PROSPECT was also utilized to generate MERIS global land products External
(i.e., the multi-biome approach). Leaf optical properties are specified Sun zenith angle d ts Deg Fixed per pixel
by four parameters: the leaf structural parameter, N (unitless); the Sensor viewing angle d to Deg Fixed per pixel
leaf chlorophyll a + b concentration, LCC (μg cm − 2); the dry matter Relative azimuth angle d phi Deg Fixed per pixel
Fraction of diffuse incoming solar skyl No dimension 0.1
content, Cm (g cm − 2); and the equivalent water thickness, Cw
radiation
(g cm − 2) (Jacquemoud et al., 2000). The top of canopy reflectance
a
simulation involves three parameters: LAI (m 2 m − 2); mean leaf incli- Coupled with equivalent water thickness in a ratio of 4:1; range in the field-level
study: 0.01–0.02.
nation angle, ALA (deg), assuming an ellipsoidal distribution of foliage b
Range in the field-level study: 0.3–7.5.
elements (Campbell, 1986); and the hot spot size parameter, hot c
Range in the field-level study: 40–70.
(m m − 1), implemented by Kuusk (1991). Geometrical parameters d
Fixed at 0, 0, and 30 in the field-level study.
 Y. Si et al. / Remote Sensing of Environment 121 (2012) 415–425 419

Table 3
R2, RMSE, and normalized RMSE (NRMSE) of grass property retrieval using different number of cases in the solution.

No. cases in Time LAI (m− 2 m− 2) LCC (ug cm− 2) CCC (g m− 2)

the solution
R2 RMSE NRMSE% R2 RMSE NRMSE% R2 RMSE NRMSE%

1 best case Apr 0.30 0.55 42 0.25 15.98 105 0.54 0.32 72
Jun 0.20 2.98 48 0.17 11.74 52 0.25 0.53 34
Pooled 0.49 2.09 34 0.20 14.13 62 0.53 0.44 28
50 best cases Apr 0.45 0.41 31 0.34 16.00 106 0.53 0.28 64
Jun 0.51 1.66 27 0.13 10.96 48 0.40 0.45 29
Pooled 0.70 1.18 19 0.17 13.85 60 0.60 0.37 24
100 best cases Apr 0.47 0.38 29 0.29 15.63 103 0.54 0.27 61
Jun 0.50 1.43 23 0.09 10.87 48 0.41 0.43 28
Pooled 0.70 1.02 16 0.14 13.59 60 0.61 0.36 23

LCC: leaf chlorophyll content; CCC: canopy chlorophyll content; RMSE: root mean square error; NRMSE: RMSE divided by the range of the parameter as measured in the field.

(i.e., from November to April). The coupling of water thickness and
dry matter content was demonstrated as a valid constraint to reduce
the ill-posed problem (Combal et al., 2002; Weiss et al., 2000). The
two parameters were coupled in a ratio of 4:1 according to Vohland
and Jarmer (2008), as this provided an improved estimation for the
experimental grassland. For the leaf structural parameter N, previous
studies have used a fixed value of 1.4 for corn and 1.7 for soybean
(Jacquemoud et al., 2000), a range of N = 2 ± 0.34 for wheat crop
(Atzberger et al., 2003), a fixed value of 1.55 for various crops
including corn, soybean, wheat and spring barley (Haboudane et al.,
2004), or a fixed value of 1.6 for grasses (Vohland & Jarmer, 2008).
Considering grasses have relatively thin leaves, we have selected a
range of N = 1.5–1.9, consistent with Darvishzadeh et al. (2008a).
The fraction of diffuse incoming solar radiation skyl depends on atmo-
spheric conditions, solar zenith angle, and wavelength. Since skyl has
relatively small influence on canopy reflectance compared to other
leaf and canopy attributes, a fixed value of 0.1 across all wavelengths
was selected, in accordance with previous studies (Darvishzadeh et
al., 2008a; Schlerf & Atzberger, 2006).
For each parameter, 100,000 values were drawn randomly within
the specific ranges and a total of 100,000 canopy reflectances were
generated based on these parameters. The number of 100,000 was
selected in accordance with Weiss et al. (2000) who suggested that
this provided a good compromise between the computer resource
requirement and the accuracy of canopy variable estimates. To select
the optimal spectra corresponding to a given measurement, the RMSE
(root mean square error) between measured and modeled spectra
was calculated. Instead of picking up the best spectrum (lowest
RMSE) and its corresponding parameters as the solution, we consid-
ered the distribution of the multiple best spectra (cases) and comput-
ed the median of their corresponding parameters. Weiss et al. (2000)
found that using the best spectrum does not guarantee to find the
best solution. Weiss et al. (2000) suggested using the 50 best cases
as the solution to retrieve LAI. Darvishzadeh et al. (2008a, 2008b)
demonstrated that the solution of the 100 best cases resulted in the
best LAI, LCC, and CCC retrievals. We tested the performance of a
different number of cases in the solution in estimating grass LAI,
LCC, and CCC. The estimated LAI, LCC, and CCC in April and June
were validated against the in-situ measurements. R 2, RMSE and
NRMSE (NRMSE = RMSE / range of the parameter as measured in the
field) were adopted to evaluate the performance of the approach.
2.8. The MERIS multi-biome products
The MERIS multi-biome products were generated using a neural
network approach. Four variables ( LAI, CCC, fraction of absorbed
photosynthetically active radiation (fAPAR), and vegetation cover
(fCover)) are estimated, using top of canopy reflectance measure-
ments in 11 spectral bands and the geometry of observation
(Bacour et al., 2006). The network was trained on a synthetic dataset
made of radiative transfer model simulations generated by the
PROSAIL model (Bacour et al., 2006). The performances of the four
retrieved variables were evaluated using the MODIS (LAI and
fAPAR) and the MERIS global vegetation index (MGVI) products on
a selection of sites corresponding to various biome types and
vegetation cycles (Baret et al., 2006). The multi-biome algorithm is
implemented in the BEAM toolbox (ESA) “TOA_VEG Processor”
(http://www.brockmann-consult.de/cms/web/beam) and was used
to estimate LAI and CCC from five MERIS L1b images. To facilitate
comparison, LCC maps were calculated by dividing the CCC maps by
the LAI maps. For each sample plot location, the values of LAI, LCC,
and CCC, as estimated by the multi-biome approach, were extracted
from the April and June maps and validated against the field
measurements.

Fig. 3. Measured and simulated MERIS reflectance of three grass sites with different LAI and canopy chlorophyll content (CCC), evaluated by RMSE (root mean square error). Dots
indicate the location of the 11 MERIS bands utilized in the analysis.
420 Y. Si et al. / Remote Sensing of Environment 121 (2012) 415–425

2.9. Temporal consistency of the multi-biome and single-biome
approaches
The agricultural and the semi-natural grasslands (including salt
marshes, herbaceous vegetation, and swamp vegetation) were extracted
from the Dutch National Landuse Database (LGN 4, 25 m spatial resolu-
tion). The extracted grassland mask was resampled to a 300 m spatial
resolution and then applied to the 5 atmospherically corrected MERIS
images. Time-series maps describing the spatio-temporal variation of
grassland quantity and quality were produced using 5 masked grassland
images as input. The mean and the standard deviation of time-series
maps generated using single-biome and multi-biome approaches were
calculated and the temporal consistency was then compared.
3. Results
3.1. Estimation of grass variables using the single-biome and multi-
biome approaches
The solution containing the best fitting case had the highest RMSE,
whereas the solution of the 50 best cases had an intermediate RMSE
and the solution of the 100 best cases yielded the lowest RMSE
(Table 3). We therefore chose the 100 best cases and the corresponding
distribution of each parameter as the solution for the retrieval. Fig. 3
shows the measured and simulated MERIS reflectance retrieved in
this way for three grass sites.
The relationships between in-situ measured grass variables and
estimates using the single-biome approach are shown in Fig. 4. A
good relationship was observed between estimated and measured
LAI using the April data, while a relatively larger RMSE was observed
using the June data. A poor relationship was observed between the
estimated and measured LCC in both single-season and pooled data
sets. The estimates of CCC showed an intermediate accuracy with a
slightly overestimation using the April data and a relatively larger
RMSE using the June data. A slightly overestimation at the lower
CCC (b1) was found using the pooled data, with an R 2 of 0.60 and a
NRMSE of 22%. Given the poor retrieval of the LCC, the accuracy of
the CCC estimate is mainly influenced by the accuracy of the LAI
estimate.
The relationships between the in-situ measured grass variables
and estimates using the multi-biome approach are shown in Fig. 5.
LAI was heavily overestimated for values smaller than 2 and LAI satu-
rated at a level around or higher than 4. Similar to the single-biome
approach, also using the multi-biome approach the retrieval of the
LCC failed. Both LCC and CCC were greatly overestimated by the
multi-biome approach using either single-season or pooled data sets.
Though both approaches failed to estimate LCC, the single-biome
approach yielded a significant improvement in estimating grass LAI

Fig. 4. In-situ measured vs. estimated grass leaf area index (LAI), leaf chlorophyll content (LCC), and canopy chlorophyll content (CCC), using the single-biome approach in April,
June, and April and June pooled. The accuracy is evaluated by the R2, the root mean square error (RMSE), and the normalized root mean square error (NRMSE) calculated by dividing
RMSE by the range of the parameter as measured in the field.
 Y. Si et al. / Remote Sensing of Environment 121 (2012) 415–425 421

and CCC compared to the multi-biome approach using either single-
season or pooled data sets (Figs. 4 and 5).
3.2. Comparison of temporal consistency of the single-biome and
multi-biome approaches
The time-series single-biome LAI maps (Fig. 6) show a low LAI in
February, an increased LAI in April, a peak LAI in June and September,
and the lowest LAI in December. The time-series single-biome CCC
(Fig. 7) maps show a similar temporal trend with the seasonal variation
of LAI. The spatio-temporal variation of grassland LAI and CCC estimated
by the single-biome and multi-biome approaches are shown in Fig. 8.
Similar trajectories were observed for both approaches, but the multi-
biome trajectory values were consistently higher. A larger spatial varia-
tion was found in single-biome LAI in September and June compared to
the other seasons.
4. Discussion
4.1. Comparison of the single- and multi-biome approaches
Both the single-biome and the multi-biome results were validated
by two-season ground truth data. Though evaluated in a relatively
heterogeneous landscape (Fig. 1), the single-biome approach still
yielded a good accuracy for LAI (R2 = 0.70, RMSE = 1.02, NRMSE=
16%) and CCC retrieval (R2 = 0.61, RMSE = 0.36, NRMSE = 23%). This
is a significant improvement compared to retrieval by the multi-
biome approach of LAI (R2 = 0.36, RMSE = 1.77, NRMSE = 28%) and
CCC (R2 = 0.47, RMSE= 1.33, NRMSE = 84%). The single-biome LAI
and field measured LAI were relatively well distributed around the
1:1 line while the single-biome CCC showed slight overestimation at
lower values (Fig. 4). However, the multi-biome LAI was overestimated
at values smaller than 2, and saturated at values equal to or larger than
4 (Fig. 5). Furthermore, the multi-biome CCC was considerably overes-
timated across the whole data range (Fig. 5). The overestimation and
saturation of LAI was observed in the original study (Bacour et al.,
2006) in which the multi-biome global product was developed. This
overestimation and saturation of LAI was also reported by a previous
study validating MERIS multi-biome LAI products across different veg-
etation species, including crops, pasture and forest (Canisius et al.,
2010). The use of different retrieval techniques in the single-biome
(LUT) and multi-biome (neural nets) approaches may partly explain
the observed differences in the retrieval. Nevertheless, it is not the ob-
jective of this paper to compare the different retrieval techniques but
to compare the accuracy of the products using field measurements as
a reference. Finally, since the estimate of LCC failed by using both

Fig. 5. In-situ measured vs. estimated grass leaf area index (LAI), leaf chlorophyll content (LCC), and canopy chlorophyll content (CCC), using the multi-biome approach in April,
June, and April and June pooled. The accuracy is evaluated by the R2, the root mean square error (RMSE), and the normalized root mean square error (NRMSE) calculated by dividing
RMSE by the range of the parameter as measured in the field.
422 Y. Si et al. / Remote Sensing of Environment 121 (2012) 415–425
Fig. 6. Time-series maps of single-biome leaf area index (LAI) (m2 m− 2) for the grassland in the northern part of The Netherlands in 2008.
approaches, the accuracy of the CCC estimate is greatly influenced by
that of the LAI estimate.
4.2. Comparison with other studies
The estimate of grass properties by the single-biome approach
showed comparable results (LAI: NRMSE = 16%, CCC: NRMSE = 23%)
with two field-level studies dealing with grasslands, i.e., Darvishzadeh
et al. (2008a) (LAI: NRMSE = 18%, CCC: NRMSE= 18%) and Vohland
and Jarmer (2008) (LAI: NRMSE= 12%, CCC: NRMSE= 13%). The
MERIS bands, covering only the visible and near-infrared spectra, pro-
vide sufficient information for the retrieval of these grass biophysical
parameters. In comparison with the multi-biome (LAI: NRMSE = 28%,
CCC: NRMSE = 84%), the single-biome approach showed a significantly
higher accuracy in estimating grass LAI (NRMSE = 16%) and CCC
(NRMSE = 23%) in this geographical region. Weiss et al. (2007) validat-
ed the LAI multi-biome global products derived from VEGETATION and
reported an NRMSE of 15% for the LAI product, which is comparable to
our single-biome result (16%). However, only a limited number of vali-
dation points for each vegetation type (e.g., three points for grass) were
utilized for validating the VEGETATION LAI product; thus, more samples
collected from more regions are needed to further evaluate the accuracy
of grass LAI retrievals.
Darvishzadeh et al. (2008a) found that the estimation accuracy was
strongly dependent on the number of grass species. In our study area,
there are two species in agricultural grassland and between two to
five species in semi-natural grassland. Given the heterogeneity of the
landscape in relation to the MERIS pixel size (Fig. 1), it was not feasible
to test the influence of species number. Sensors with finer spatial
resolution could be used to test whether the estimation in agricultural
grassland has a higher accuracy than in semi-natural grassland.
4.3. Retrieval of grassland properties from two seasons by the
single-biome approach
The single-biome approach showed a better performance in esti-
mating the spring grassland (April) than the summer grassland
(June). The complex grass canopy condition in summer may have
caused this poorer retrieval, as grasslands were frequently mowed
in summer by the local farmers, sometimes with dry leaf materials
left covering the vegetation canopy. Grasses tend to grow faster
after mowing, taking advantage of the warmer temperature and suf-
ficient precipitation in the study area in summer. After mowing, dou-
ble grass layers (i.e., dried stems and flushed young grasses) were
observed during the June field campaign. Thus in future studies, the
contribution of non-green materials to the canopy spectra needs to
be taken into account (He & Mui, 2010). A background combining re-
flectance from soil, grass litter, and dried stems could be configured in
the LUT for more realistic background representation.
The scale mismatch between MERIS and the in-situ data did im-
pact our results, especially because the study area is composed of
many mixed pixels (Fig. 1). The mismatch of the sampling date and
the image capture date may also have influenced the retrieval accura-
cy, especially for the warm seasons, when grass grows faster. The
change of observation geometry of images captured on different
dates may have caused the change of the heterogeneity level of
each MERIS pixel (Gomez-Chova et al., 2010), which may also have
led to different accuracies in different seasons. Additionally, cloud
 Y. Si et al. / Remote Sensing of Environment 121 (2012) 415–425
Fig. 7. Time-series maps of single-biome canopy chlorophyll content (CCC) (g m− 2) for the grassland in the northern part of The Netherlands in 2008.
contaminated pixels may bias the retrieval of grass properties from
MERIS measurements. Although we have used an updated cloud fil-
tering technique coupled to an atmospheric correction process
(Guanter et al., 2008), the influence of thin clouds was still observed
in the generated time-series grassland maps (the stripes observed in
the June image in Fig. 6). Further efforts should be taken to solve this
cloud mask issue before accurate maps can be produced and used
(Bacour et al., 2006). Finally, the use of effective LAI may partly ex-
plain deviations between the retrieved true LAI and field measure-
ments (which tend to be underestimated).
4.4. LUT configuration
The LUT approach is ill-posed as various combinations of canopy
parameters may yield similar spectra. Prior information can be useful
for constraining retrievals. However, specificity to local conditions
limits the applicability to other regions. In our case, we make maxi-
mal use of known information to constrain the parameter range. We
fixed the view angle, the zenith angle, and the relative azimuth
angle, and we also limited the LAI range based on the phenological
information. These constraints/settings can be adjusted based on
regional conditions so that the LUT can be applied to other regions.
Meanwhile, we selected relatively broad ranges for LAI, LCC, and
leaf angle to keep our approach generic and we chose an N range
appropriate for retrieving grass properties. Clearly, any LUT
configuration must deal with the trade-off between generic and
local parameterizations. Still, we believe that our single-biome
approach can be applied to other sites or at larger scales by
423
incorporating different soil backgrounds, per-pixel geometrical infor-
mation and phenology-based LAI ranges.
4.5. Retrieval of LCC at the MERIS level
The estimate of LCC failed at the MERIS FR level with both the
single- and the multi-biome approaches. One possible reason could
be the heterogeneous nature of the surfaces at this medium spatial
resolution—Fig. 1 shows that the study area has many mixed pixels.
One possible alternative to overcome the heterogeneity problem is
to make use of data fusion techniques (Zurita-Milla et al., 2009). Nev-
ertheless, the LCC failure at the MERIS level was expected, as difficul-
ties in estimating LCC for grasslands have been already reported at
the field level (Darvishzadeh et al., 2008a). The poor signal propaga-
tion from leaf to canopy has been stated by several studies (Asner,
1998; Yoder & Pettigrew-Crosby, 1995; Zarco-Tejada et al., 2004).
Moreover, He and Mui (2010) suggested that the contribution of
non-green materials from canopy spectra in grasslands makes it
more difficult to retrieve biochemical content from remote sensing
data at the canopy level. In order to check whether our LCC results
from LUT inversion were reliable, we pooled the field data and pre-
dicted LCC using 11 MERIS bands and the partial least square (PLS) re-
gression. Similar results were observed for the single-biome approach
(R 2 = 0.14, RMSE = 12.59, NRMSE = 60%) and for the PLS regression
(R 2 = 0.31, RMSE = 12.02, NRMSE = 53%). Furthermore, the frequent
fertilization in the study area may lead to a relatively small range of
LCC thereby increasing the difficulty of the LCC retrieval. Meanwhile,
the poor LCC retrieval from the multi-biome approach could be due to
424 Y. Si et al. / Remote Sensing of Environment 121 (2012) 415–425
Fig. 8. Temporal variation of grassland leaf area index (LAI m2 m–2) (a) and canopy
chlorophyll content (CCCg m–2) (b) in the northern Netherlands in 2008 estimated
by the single-biome (solid bars) and the multi-biome approach (open bars), showing
mean and standard deviation.
the fact that it was calculated by simply dividing CCC by LAI, which
could not be reliably estimated by this approach.
4.6. Evaluation of the temporal consistency
The time-series grassland LAI and CCC maps estimated by the
single-biome approach are highly associated with grassland phenology
and are also influenced by agricultural practices. The relatively large
spatial variation in June and September maps is caused by mowing
activity. As frequent fertilization leads to a consistently high quality
grass all year round, the temporal variation of CCC is mostly influenced
by the variation of LAI. The single-biome temporal trajectory is consis-
tent with the multi-biome trajectory, showing the phenological
development of grasses. Because of the overestimation of the multi-
biome LAI, the multi-biome trajectory values were consistently larger
than the single-biome trajectory values (Fig. 8). Besides, the larger
standard deviation of June and September LAI estimated by the
single-biome approach can be explained by the coexistence of
mowed and unmown grasses in the field.
5. Conclusions
This study proposes a PROSAIL-based single-biome approach to es-
timate the quantity and quality of grassland at the MERIS level. The
proposed single-biome approach was compared with the operational
MERIS global multi-biome products (i.e., the multi-biome approach).
By using two-season ground data, we quantified how much better
the proposed approach is. The following conclusions are drawn from
this study: 1) the single-biome approach (LAI: NRMSE = 16%, CCC:
NRMSE = 23%) yields significantly higher accuracy than the multi-
biome approach (LAI: NRMSE = 28%, CCC: NRMSE= 84%) in estimating
grassland properties in the heterogeneous study area; 2) the retrieval
of LCC fails at the MERIS (FR) medium spatial resolution irrespective
of the retrieval approach; and 3) the spatio-temporal patterns of LAI
and CCC estimated by the single-biome approach are consistent with
the phenology of grass and at the same time they show the influence
of mowing and fertilizing practices. The proposed method is suitable
for producing regional time-series LAI and CCC maps. These important
vegetation properties are helpful to understand, for instance, the
spatio-temporal variation of forage availability for wild birds. Inconsis-
tencies between satellite maps and field data could be further reduced
by fine tuning the inversion routine, reducing the heterogeneity prob-
lem, enhancing atmospheric correction accuracy and improving the
match between image acquisition and field data collection.
Acknowledgement
This research is sponsored by the China Scholarship Council (CSC)
and the Faculty of Geo-Information Science and Earth Observation
(ITC), University of Twente, The Netherlands. We are grateful to
Luis Guanter (GFZ, Potsdam, Germany) for assistance in the atmo-
spheric correction of MERIS FR L1b images and Willem Nieuwenhuis
for programming the calculation routines. We thank Wouter Verhoef
and three anonymous reviewers for their insightful comments and
suggestions.
Appendix A. Supplementary data
Supplementary data to this article can be found online at doi:10.
1016/j.rse.2012.02.011.
References
Asner, G. P. (1998). Biophysical and biochemical sources of variability in canopy
reflectance. Remote Sensing of Environment, 64, 234–253.
Atzberger, C. (2004). Object-based retrieval of biophysical canopy variables using
artificial neural nets and radiative transfer models. Remote Sensing of Environment,
93, 53–67.
Atzberger, C., Jarmer, T., Schlerf, M., Kotz, B., & Werner, W. (2003). Retrieval of wheat
bio-physical attributes from hyperspectral data and SAILH + PROSPECT radiative
transfer model. In M. Habermeyer, A. Muller, & S. Holzwarth (Eds.), 3rd EARSeL
workshop on imaging spectroscopy. Germany: Herrsching.
Bacour, C., Baret, F., Beal, D., Weiss, M., & Pavageau, K. (2006). Neural network estima-
tion of LAI, fAPAR, fCover and LAIxCab, from top of canopy MERIS reflectance data:
Principles and validation. Remote Sensing of Environment, 105, 313–325.
Baret, F., Bacour, C., Weiss, M., & Berthelot, B. (2006). Report on the validation of MERIS
TOC_VEG land products. : European Space Agency.
Baret, F., & Guyot, G. (1991). Potentials and limits of vegetation indices for LAI and
APAR assessment. Remote Sensing of Environment, 35, 161–173.
Baret, F., Weiss, M., Troufleau, D., Prevot, L., & Combal, B. (2000). Maximum informa-
tion exploitation for canopy characterization by remote sensing. Aspects of Applied
Biology, 60, 71–82.
Botha, E. J., Leblon, B., Zebarth, B. J., & Watmough, J. (2010). Non-destructive estimation
of wheat leaf chlorophyll content from hyperspectral measurements through
analytical model inversion. International Journal of Remote Sensing, 31, 1679–1697.
Campbell, G. S. (1986). Extinction coefficients for radiation in plant canopies calculated
using an ellipsoidal inclination angle distribution. Agricultural and Forest Meteorology,
36, 317–321.
Canisius, F., Fernandes, R., & Chen, J. (2010). Comparison and evaluation of Medium
Resolution Imaging Spectrometer leaf area index products across a range of land
use. Remote Sensing of Environment, 114, 950–960.
Clevers, J. G. P. W., Schaepman, M. E., Mucher, C. A., Wit, A. J. W. D., Zurita-Milla, R., &
Bartholomeus, H. M. (2007). Using MERIS on Envisat for land cover mapping in
the Netherlands. International Journal of Remote Sensing, 28, 637–652.
Combal, B., Baret, F., Weiss, M., Trubuil, A., Mace, D., Pragnere, A., et al. (2002). Retrieval
of canopy biophysical variables from bidirectional reflectance: using prior
information to solve the ill-posed inverse problem. Remote Sensing of Environment,
84, 1–15.
Darvishzadeh, R., Skidmore, A., Schlerf, M., & Atzberger, C. (2008). Inversion of a radiative
transfer model for estimating vegetation LAI and chlorophyll in a heterogeneous
grassland. Remote Sensing of Environment, 112, 2592–2604.
Darvishzadeh, R., Skidmore, A., Schlerf, M., Atzberger, C., Corsi, F., & Cho, M. (2008). LAI
and chlorophyll estimation for a heterogeneous grassland using hyperspectral
measurements. ISPRS Journal of Photogrammetry and Remote Sensing, 63, 409–426.
Dorigo, W. A., Zurita-Milla, R., de Wit, A. J. W., Brazile, J., Singh, R., & Schaepman, M. E.
(2007). A review on reflective remote sensing and data assimilation techniques for
 Y. Si et al. / Remote Sensing of Environment 121 (2012) 415–425
enhanced agroecosystem modeling. International Journal of Applied Earth
Observation and Geoinformation, 9, 165–193.
Filella, L., Serrano, L., Serra, J., & Penuelas, J. (1995). Evaluation of wheat crop nitrogen
status by remote sensing: Reflectance indices and discriminant analysis. Crop
Science, 35, 1400–1405.
Gobron, N., Pinty, B., & Verstraete, M. M. (1997). Theoretical limits to the estimation of
the leaf area index on the basis of visible and near-infrared remote sensing data.
IEEE Transactions on Geoscience and Remote Sensing, 35, 1438–1445.
Gomez-Chova, L., Zurita-Milla, R., Alonso, L., Amoros-Lopez, J., Guanter, L., &
Camps-Valls, G. (2010). Gridding artifacts on medium-resolution satellite image
time series: MERIS case study. IEEE Transactions on Geosciences and Remote Sensing,
49, 2601–2611.
Guanter, L., Gomez-Chova, L., & Moreno, J. (2008). Coupled retrieval of aerosol optical
thickness, columnar water vapor and surface reflectance maps from ENVISAT/
MERIS data over land. Remote Sensing of Environment, 112, 2898–2931.
He, Y., & Mui, A. (2010). Scaling up semi-arid grassland biochemical content from the
leaf to the canopy level: Challenges and opportunities. Sensors, 10, 11072–11087.
Haboudane, D., Miller, J. R., Pattey, E., Zarco-Tejada, P. J., & Strachan, I. B. (2004). Hyper-
spectral vegetation indices and novel algorithms for predicting green LAI of crop
canopies: Modeling and validation in the context of precision agriculture. Remote
Sensing of Environment, 90, 337–352.
Houborg, R., Soegaard, H., & Boegh, E. (2007). Combining vegetation index and model
inversion methods for the extraction of key vegetation biophysical parameters
using Terra and Aqua MODIS reflectance data. Remote Sensing of Environment,
106, 39–58.
Horler, D. N. H., Dockray, M., & Barber, J. (1983). The red edge of plant leaf relfectance.
International Journal of Remote Sensing, 4, 273–288.
Jacquemoud, S., Bacour, C., Poilve, H., & Frangi, J. -P. (2000). Comparison of four radia-
tive transfer models to simulate plant canopies reflectance: Direct and inverse
mode. Remote Sensing of Environment, 74, 471–481.
Jacquemoud, S., & Baret, F. (1990). PROSPECT: a model of leaf optical properties
spectra. Remote Sensing of Environment, 34, 75–91.
Jacquemoud, S. e., Verhoef, W., Baret, F., Bacour, C., Zarco-Tejada, P. J., Asner, G. P., et al.
(2009). PROSPECT+SAIL models: a review of use for vegetation characterization.
Remote Sensing of Environment, 113, S56–S66.
Jenkins, G. I., Baker, N. R., & Woolhouse, H. W. (1981). Changes in chlorophyll content
and organization during senescence of the primary leaves of Phaseolus vulgaris L. In
relation to photosynthetic electron transport. Journal of Experimental Botany, 32,
1009–1020.
Kimes, D. S., Knyazikhin, Y., Privette, J. L., Abuelgasim, A. A., & Gao, F. (2000). Inversion
methods for physically-based models. Remote Sensing Reviews, 18, 381–439.
Kokaly, R. F., Asner, G. P., Ollinger, S. V., Martin, M. E., & Wessman, C. A. (2009).
Characterizing canopy biochemistry from imaging spectroscopy and its application
to ecosystem studies. Remote Sensing of Environment, 113, S78–S91.
Kuusk, A. (1991). The hot-spot effect in plant canopy reflectance. In R. B. Myneni, & J.
Ross (Eds.), Photon–vegetation interactions. New York: Springer-Verlag.
Maire, G. L., Marsden, C., Verhoef, W., Ponzoni, F. J., Seen, D. L., Begue, A., et al. (2011).
Leaf area index estimation with MODIS reflectance time series and model
inversion during full rotations of Eucalyptus plantations. Remote Sensing of
Environment, 115, 586–599.
425
Markwell, J., Osterman, J. C., & Mitchell, J. L. (1995). Calibration of the Minolta
SPAD-502 leaf chlorophyll meter. Photosynthesis Research, 46, 467–472.
Medina, E., & Lieth, H. (1963). The cholorphyll content of some plant associations in
Central Europe and its relationship to the productivity. Qualitas Plantarum et
Materiae Vegetabiles, 9, 217–229.
Mutanga, O., Skidmore, A. K., & Prins, H. H. T. (2004). Predicting in situ pasture quality
in the Kruger National Park, South Africa, using continuum-removed absorption
features. Remote Sensing of Environment, 89, 393–408.
Rast, M., Bezy, J. L., & Bruzzi, S. (1999). The ESA Medium Resolution Imaging Spectrom-
eter (MERIS): a review of the instrument and its mission. International Journal of
Remote Sensing, 20, 1681–1702.
Savitzky, A., & Golay, M. J. E. (1964). Smoothing and differentiation of data by
simplified least square procedure. Analytical Chemistry, 36, 1627–1638.
Schlerf, M., & Atzberger, C. (2006). Inversion of a forest reflectance model to estimate
structural canopy variables from hyperspectral remote sensing data. Remote
Sensing of Environment, 100, 281–294.
van der Meer, F., Bakker, W., Scholte, K., Skidmore, A., Jong, S. D., Clevers, J., et al.
(2001). Spatial scale variations in vegetation indices and above-ground biomass
estimates: Implications for MERIS. International Journal of Remote Sensing, 22,
3381–3396.
Verhoef, W. (1984). Light scattering by leaf layers with application to canopy
reflectance modeling: the SAIL model. Remote Sensing of Environment, 16, 125–141.
Verhoef, W. (1985). Earth observation modeling based on layer scattering matrices.
Remote Sensing of Environment, 17, 165–178.
Vohland, M., & Jarmer, T. (2008). Estimating structural and biochemical parameters for
grassland from spectroradiometer data by radiative transfer modelling (PROSPECT
+SAIL). International Journal of Remote Sensing, 29, 191–209.
Weiss, M., & Baret, F. (1999). Evaluation of canopy biophysical variable retrieval
performances from the accumulation of large swath satellite data. Remote Sensing
of Environment, 70, 293–306.
Weiss, M., Baret, F., Garrigues, S., & Lacaze, R. (2007). LAI and fAPAR CYCLOPES global
products derived from VEGETATION. Part 2: Validation and comparison with
MODIS collection 4 products. Remote Sensing of Environment, 110, 317–331.
Weiss, M., Baret, F., Myneni, R. M., Pragnere, A., & Knyazikhin, Y. (2000). Investigation
of a model inversion technique to estimate canopy biophysical variables from
spectral and directional reflectance data. Agronomie, 20, 3–22.
Wellburn, A. R. (1994). The spectral determination of chlorophylls a and b, as well as
total carotenoids, using various solvents with spectrophotometers of different res-
olution. Journal of Plant Physiology, 144, 307–313.
Wright, I. J., Reich, P. B., Westoby, M., Ackerly, D. D., Baruch, Z., Bongers, F., et al. (2004).
The worldwide leaf economics spectrum. Nature, 482, 821–827.
Yoder, B. J., & Pettigrew-Crosby, R. E. (1995). Predicting nitrogen and chlorophyll
content and concentrations from reflectance spectra (400–2500 nm) at leaf and
canopy scales. Remote Sensing of Environment, 53, 199–211.
Zarco-Tejada, P. J., Miller, J. R., Morales, A., Berjon, A., & Aguera, J. (2004). Hyperspectral
indices and model simulation for chlorophyll estimation in open-canopy tree
crops. Remote Sensing of Environment, 90, 463–476.
Zurita-Milla, R., Kaiser, G., Clevers, J. G. P. W., Schneider, W., & Schaepman, M. E. (2009).
Downscaling time series of MERIS full resolution data to monitor vegetation
seasonal dynamics. Remote Sensing of Environment, 113, 1874–1885.