International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part Bl. Istanbul 2004
continental biomes. For almost all ecosystems, the POLDER-1
and the POLDER-2 NDVI present a full relevant vegetation
cycle. During the northern springtime, when POLDER-1 and
POLDER-2 synthesis are available, we can note the differences
between the 2 years: in 2003 (POLDER-2) the vegetation
growth started some weeks earlier than in 1997 (POLDER-1).
This is due to the fact that spring 2003 was warm before a very
hot summer. These profiles are regular and smooth what
translate a good temporal consistency. Note that NDVI value
equal to 0 on sub-polar crop site is due to a single observation
contaminated by a cloud. On figure 4, the FVC maps over
Africa from POLDER-1 and POLDER-2 measurements display
similar spatial features. The main differences appear on the
South-West of Africa. In Zambia and Eastern Angola, the FVC
is around 0.5 in 1997 and around 0.2 in 2003. The low
vegetation development is a consequence of the drought that
occurred in this part of Africa in 2003.
FVC
Figure 4: Fraction of Vegetation Cover over Africa retrieved in
April 1997 from POLDER-I (at left) and in April
2003 from POLDER-2 (at right).
The third step of the validation plan consists in comparing
POLDER-2 parameters with concomitant products derived from
other sensors using similar or different approaches. Then, we
performed a comparison between the POLDER-2 DHR and LAI
and the MODIS black-sky albedo and LAI.
The methodology applied to retrieve albedos from MODIS
observations is similar to the POLDER-2 approach, i.e. based
upon the inversion of a kernel-driven, linear BRDF model.
Thus, the operational MODIS BRDF/albedo algorithm makes
use of the RossThickLiSparse-Reciprocal kernel combination to
normalize MODIS data and to calculate the black-sky and
white-sky albedos (Lutch et al., 2000). The black-sky albedo is
computed for the local noon solar zenith angle for each location.
These parameters are provided at the global scale for each 16-
day period. We use global black-sky albedo provided in a CGM
grid at 0.05° resolution.
More than the differences in the sensors spectral characteristics,
the period of synthesis (16 days for MODIS, 30 days for
POLDER-2) will imply variations on the parameters, especially
when the surface conditions vary quickly, for instance because
of snow melting or snow fall. In order to reduce these effects,
we relate the POLDER-2 DHR at 565nm, 670nm and 865nm
with MODIS black-sky albedo for bands b4 (545-565nm), bl
(620-670nm), and b2 (841-876nm), respectively, over stable
desert sites. The profiles are very close with some absolute low
differences, around 196 at 865nm and less than 3% in the visible
bands. That proves the good consistency of MODIS and
POLDER-2 measurements.
The MODIS LAI retrieval algorithm relies on a main procedure
which takes advantage of the spectral and angular of the sensor.
If this algorithm fails, a back-up procedure is applied to
estimate the LAI from vegetation indices (Myneni et al., 1997).
The main algorithm is based upon a 3D radiative transfer model
which depends on the vegetation structural properties. The
measured and simulated bi-directionnal reflectances are
compared using a look-up table (Knyazikhin et al., 1998). Here,
we use the monthly MODIS LAI available on the Boston
University ftp site (wwwl).
Arctic Tundra Prutriz Wetiend Woodland
si = 1 |
| & i ] * +
| E Fol a net %
2 4, : ? +4 x
m UI X "eua 1 sp a TS
al ge em Hed n din
T "4 TWEET TEE RS TW KU Tu SEU
Temperate Fallow Grassten Tales Prairie
Fc ent pepe (EA rE pepe
af 5 + s i
i dec * *
4] # s P +
Ba 4 mt 1
ai E + 1 x wt x
st amv t He : ? 4
i eis ges i" 1 a :
FEW ITE aa VS J FNMA ETF aa ND FFWVLN/TITESRNT
Mediterranean Crops Temperate Crops Great Philos Crops
E T 1 -
za! ee © ng &
15 "d & ha 3 = 25
is " ’ - * "
5 A 1 ctt ? *. KR
at 2.78 2 t WE T * ! e rt ^ut
ee. Bt ad - put a
£a T. d of 4 e alt* M t
4? MAMJAIÀ 4G Nd JFMR JA FOND JFMARAJIREGARASGO
Deciuous Forest Deckiunos Forest Dec kduous Fores:
a te t f
5 i 4 A re a4 E 3
" 5 f P ve 5 du A
« grim Fu TH JS 1
1 £ iw. Y 3 ? 3 £5 89789 es
it = * X - 4 3 ; 5 a
? jT $ 7 y + s Thay * ^.
1 n ets ; tT x a A Leur 24
P MI * y dug
lec JI 4 ‘ a
JATHERJIA CN JP Ma RJ IAFEND JFMAMJJIARSONU
Luides Foreut Semdec uous Rulo Forest Primary Ralo Forest
? i ar Em
ü & e 07 k. E
5 poe ce Vl poorer ort
t : + m thy
y DU Spe
3 * * & vif 4 % *
? > : 7 ?
1 @ +
a 5 a
4FPFMAK&MJIASO N41 dl FMa@1 JR 80ND SFMANU IA FLN TO
Evergreen Erosilioa! Poreut Open Sauanaa Woody Havana
NEE yy 3 1 epe
5 . :
got s as Wa
1 . * il
3 Far RE s as FRY
* ibt (+ 1 a ^
4 dh a+ A £ +
1 ; >> ga se n
nes, ad n
x 3 ï ; 3
CU TS J FMEWJTRESNE sn TS RU
Figure 5: Time profiles of LAI from POLDER (magenta) and
MODIS (green) (8km resolution) over some sites
characterizing the main continental biomes. Black
symbols represent ground measurements collected
during previous years.
For most of the sites, the POLDER-2 and MODIS LAI display
similar profiles, with difference in magnitude (Figure 5).
POLDER-2 shows greater LAI over tallgrass prairie,
mediterranean crops, great plain crops, and deciduous forest.
Over arctic tundra, it is obviously too large. Over woodland,
and temperate fallow, MODIS LAI exhibits much higher values
than POLDER-2 and also than ground measurements. In
general, POLDER-2 and MODIS LAI profiles suit the time
evolution of in-situ data, except for grassland and great plains
crops. This is the consequence of the time shift between the year
of ground measurement collection and 2003.The profile of the
Landes forest of satellite product is clearly controlled by the
underwood growth. The lack of POLDER-2 data over open
Interna
savann
these c
over sit
The f:
compai
generat
VALE]
Instrun
sites w
under [
The Le
data ar
to esti
and up
to calc
module
contam
Directi
generat
tempor
POLDI
produci
Consid
parame
spatial
well 1
discrep
with «
referent
Bréon,
Analys
space. .
Calvet,
Cabelg
interact
contras!
Chen, jJ
non-fla
Descha
Bricauc
instrum
Geosci.
Hender
scheme
and het
Jacquer
Andreo
using t
Sens. Ei
Knyazil
SW. |