1121 
id a Soil Respiration 
We used monthly climatologies for solar radiation and temperature data. Indeed, we assumed that interannual 
variations in solar radiation and temperature are small compared to changes in remotely sensed vegetation 
indices, which integrate well the effects of changes in rainfall (water availability is the first determinant of 
plant growth world-wide). All input data sets have been interpolated to the weekly time resolution of the 
remotely sensed data we used. All sets have been registered to Plate Carrée projection, with a l°xl° resolution, 
from 75°N to 55°S, and 180°W to 180°E. 
:rs NPP as the product of 
lal model for crop growth 
es by the incident global 
2.2.1 NDVl and its processing 
In this study, we used the second generation of NOAA/AVHRR GVI product (Tarpley et ai, 1984) from 1986 
to 1991 (NOAA 9 from 1st January 1986 to 7 November 1988 - NOAA 11 from 8th November 1988 to 31st 
December 1991). This product provides raw digital counts in visible and near infrared spectral bands on a 
weekly basis at the 1/7° x 1/7° resolution, as a result of a cloud screening technique based on the maximum 
(1) 
NDVl of the week (Holben 1986). Using the temporal drift of sensor calibrations estimated by Kaufman and 
Holben (1993), we first computed top of the atmosphere reflectance in both channel, at the GVI resolution. The 
dive Radiation (PAR) and 
calibration is based on temporal interpolation of 13 gain coefficients from April 1985 to December 1992. 
This results in calculated NDV1 T0A , which are much less variable interannually than the raw NDVl TOA because 
calibration coefficients have variations up to 15% for each channel, NOAA 9 leading to the most variable 
:nt PAR) 
reflectances. Then, we applied a method for correction of atmospheric effects based on a parameterization of 
the basic equations describing the essential processes of radiative transfer in the atmosphere (Rahman and 
ed Normalized Difference 
Dedieu 1994) which accounts for molecular and aerosol scattering, and also for ozone and water vapour 
transmission. Water vapour and ozone contents were derived from climatologies (Oort 1983),and (London 
(2) 
1976) . The atmospheric optical depth has been set to 0.05 at 550 nm. Correction of scattering and absorption 
effects takes into account the altitude through mean vertical profiles (Berthelot 94). NDVl SURF is then derived 
after processing the reflectances at the l°xl° resolution. 
s of study (0 £ f(t) £ 1). In 
is we focused on 1989, we 
TOA , a second one for the 
the text, TOA and SURF 
m. 
re consider that, at the first 
2.2.2 Other inputs 
♦ Temperature is extracted from the Lee mans and Cramer (1991) database, which provides monthly mean 
temperatures for 30 years. 
♦ Incoming global solar radiation at the surface is derived on a monthly basis from an output of GCM of the 
French Meteorological Office (Planton et al 1991). Satellite based global solar radiation maps are currently 
being processed by NASA Langley Research Center for WCRP/GEWEX (Whitlock et al. 1993). 
♦ Atmospheric CO2 measurements are taken from the Carbon Dioxide Information Analysis Center (CDLAC). 
(3) 
milation)/ (absorbed PAR) 
rophic respiration, 
canopies against absorbed 
it regression between these 
ninary study, we used the 
on temperature only : 
3 - RESULTS AND DISCUSSION 
We have ran our model from 1986 to 1991, using the NOAA-AVHRR derived NDVf rOA for these years. In our 
analysis of results, we tried to distinguish between measurements uncertainties (correction of calibration 
changes, orbital drift of the satellite, changes in atmospheric composition after the Pinatubo eruption), and 
variations connected to actual changes of the vegetation. Then, we focused on the year 1989 to which we 
applied atmospheric correction. This way, we assess the sensitivity of our model to the preliminary atmospheric 
(4) 
decontamination. Indeed, when the model is run by increasing by 1% NDVl TOA , the annual NPP of the globe is 
increased by 1.4% (roughly 1 Gt(C)). So, we presented the impact of correction at three main stages of the 
modelling that we selected as being GVI, NPP and NEP. This approach allow us to emphasize advantages and 
lacks accompanying the use of GVI (and their further processing) as "pilot" input for NEP models. 
id soil-related parameters : 
temperature. In this model, 
3.1 GVI 
We synthesized the whole GVI data set from 1986 to 1991, through a "latitude- time diagram" including the six 
(5) 
years. Fig.l presents this result with a weekly, zonally cumulated GVI, from January 1986 to December 1991. 
Each column presents the weekly time step, and one line corresponds to a 1° latitude zonal band. A point of the 
diagram is so the sum of GVI for terrestrial pixels on the considered latitude for one date. Within 1° belt, for 
one date, we did not weigh GVI by the corresponding surface of land in order to preserve latitudinal 
information 
rature T ( Qjq = 1.5 in this 
librium over a year on each 
>e improved in the future by 
3.1.1 Artefacts due to satellite data: 
• In the GVI data set, we distinguish problems involved with radiometer calibration, and those linked to an 
abrupt and sizeable atmospheric event Indeed, we can observe a low signal decrease from 1986 to 1987 with a 
stronger trend in 1988. As we accounted for the radiometer gain drift, this fact is more probably due to orbital 
drift of the satellite especially in 1988. Also visible on Fig. 1 is the change from NOAA-9 to NOAA-11 by early