1123 
different conditions of 
months following the 
is built from the NDVf TOA/SURF thresholds which correspond to the 2% and 98% population levels of the 
cumulated histogram. Therefore, A and B parameters of Eq.2 are different for NDVÏ rOA and NDVI SUI ^ F . 
te middle of 1991 they 
e explanation for this 
, 1991). In June 1991, 
the inter tropical zone, 
i of NDVf* )A for the 
the climate conditions 
of the growing season 
ms. 
laximum in December) 
ase with the temperate 
highly in opposition of 
email number of years 
ut in the years 1986 to 
ling orbit This orbital 
ig signal, and different 
climatic event occurred 
isode in early 1991, and 
n the uptake of CO2 by 
while Pinatubo aerosols 
inatubo aerosols disturb 
1 GVI in 1989. We can 
lects in most part of the 
seated in the northern 
3.3 NER 
3.3.1 Estimated NEP time space distribution 
Fig.3 represents time evolution of cumulated, zonally averaged NEP estimates, from January 1986 to December 
1991, for 1° zonal bands and weekly time-step. When this cumulated NEP is positive (dark values in Fig.3), the 
terrestrial biosphere acts as a sink for atmospheric carbon, while negative cumulated NEP (bright values in 
Fig.3) corresponds to a release of carbon to the atmosphere. For each year we can observe, for each hemisphere, 
several zones with opposite seasonal cycles. This diagram reads this way : if the peak of NPP occurs in 
summer, the cumulated NEP is negative in the first half of the year and positive after ; the differences in 
amplitude reflect the strength of the seasonality : a zone with a marked growing season and a dormant season 
has two peaks of high amplitude, while a zone with sempervirent vegetation has small peaks to no peak at all. 
In the northern hemisphere, as we go southward, we notice three zones with opposite phases : the 
boreal/temperate with a growing season peak in summer (cold deciduous), the subtropics (monsoon in Asia) 
with a peak in winter, the tropics (savannas of Africa, South America) with a low-amplitude peak in autumn. 
At the equator there is almost no seasonality. In the southern hemisphere, we notice 2 zones with opposite 
phases : the tropics (savannas of Africa, South America) with a peak in spring, the subtropics/temperate zones 
(South America) with a peak in winter. 
3.3.2 Estimated NEP and atmospheric correction - comparison with atmospheric C02 concentration 
Even though the terrestrial biosphere is not the only compartment exchanging carbon with the atmosphere, and 
even though mixing occurs in the atmosphere, we can assume that the first phenomenon driving the seasonal 
variations of CO2 in the atmosphere as measured in CO2 sampling stations is the terrestrial NEP, at least for 
the northern hemisphere (Heimann et al. 1989). We can therefore improve the approach of Tucker et al. (1986) 
through a quantitative comparison between time-integrated NEP fluxes computed with calibrated NDVI, and 
atmospheric CO2 concentrations ([CO2]), instead of directly relating satellite-derived vegetation indices to 
[СО2]. For example, we plot Fig.7-a the temporal evolution of [CO2] (sampled at Niwot Ridge 40 0 N) as a 
function of time, assuming that a station is representative of the 1° belt where it is located. Simultaneously, we 
represent on the same figure the evolution of estimated cumulated NEP (curve b) and NEP T0A (curve a) for a 
1° zonal band. The relationship between [C02] and NEP displayed there, is representative of the mean trend 
observed in northern hemisphere. It suggest that in that region, NEP is the primary driver of [С02]. 
Differences between curve(a) and curve (b) show that for this latitude atmospheric correction reduce the NEP 
estimation. Figure 7b present this relative difference, with a reduction of 10 % in summer. This figure can be 
explained as being an extraction of the figure 6 which also represents the relative impact of atmospheric 
correction on NEP, all over the considered latitudinal rang. We remind that dark values represent a decrease on 
estimations, induced by atmospheric decontamination. 
t appears that the mean 
ugh peak in summer for 
and 30°N, and the inter- 
nisphere peak, 
r model (unit: 10 15 gof 
1991 
96.7 
67.7 
lified spatial distribution, 
• rall y the estimated NPP, 
the difference (M > / >T0A * 
and zones for which the 
yfrOA/SURF and/rt) (Eq.2) 
4 - CONCLUSION 
Our ability to analyse interannual variation of NPP and NEP, strongly depends on the accuracy of NDVI. This 
study emphasizes some tricky points about estimation artefacts involved with satellite measurements. 
As a matter of fact, when the model is ran by increasing by 1% NDVI*, the annual NPP of the globe is 
increased by 1.4% (roughly 1 Gt(C)), and we saw that accounting for atmospheric effects can induce a rise of 
6% in estimated NPP. The example of 1988 shows us that we will not be able to use this year as input without 
correcting the NOAA 9 orbital drift We also need to account for Pinatubo aerosols in the 1991 GVI data set 
More generally we have to improve each input data set especially for optical depth which was set to a constant 
low value in this experiment. If we consider the way atmospheric correction acts upon estimated NEP, we can 
observe fluctuations ran g in g from a 15% decrease to a 25% increase. 
Considering this large sensitivity with regard to the quality of remotely sensed signal, we certainly have to 
improve the GVI processing by (for example) accounting for surface directional effects. Al this stage of the 
analysis, global NEP estimation should be used in a transport model, in order to validate and calibrate our 
terrestrial biosphere model, driven by an accurate GVI data set 
acknowledgements 
This work is a contribution to the ESCOBA project "The global carbon cycle and its perturbation by man and 
climate n. Part B : terrestrial biosphere", supported by the Environment Program of the Commission of the 
European Communities. The authors would like to thank also the "Region Midi-Pyrenees" for funding part of