The Sahel region has been identified as a “hot spot” of global environmental change, but understanding of the roles of different climatic and anthropogenic forcing factors driving change in the region is incomplete. We show that a process-based ecosystem model driven by climatic and atmospheric CO2 data alone closely reproduces the satellite-observed greening trend of the Sahel vegetation and its interannual variability between 1982 and 1998. Changes in precipitation were identified as the primary driver of the aggregated simulated vegetation changes. According to the model, the increasing carbon uptake through vegetation was associated with an increasing relative carbon sink; but integrated over the whole period, the Sahel was predicted to be a net source of carbon.
 The Sahel belt of Africa has been the focus of much scientific attention in recent decades. Mean annual rainfall amounts have been consistently below normal since the late-1960s [Hulme, 2001; Foley et al., 2003], causing recurring drought and famine in the region [Olsson, 1993]. Modeling experiments indicate that this shift towards drier conditions may have been triggered by increasing sea surface temperatures, which weaken the African monsoon, re-enforced by feedbacks of changing vegetation cover on the atmosphere [Zeng et al., 1999; Giannini et al., 2003]. The vegetation changes may be climatically caused, but may also have been affected by human-induced land degradation (via overgrazing and changed land management practices [Charney, 1975; Wang and Eltahir, 2000; Taylor et al., 2002; Foley et al., 2003]). Such land degradation might thereby lead to a catastrophic reduction in the carrying capacity of ecosystems in the region [Zeng, 2003].
 Recently, Eklundh and Olsson  showed that vegetation greenness, as estimated from satellite measurements using the Normalized Difference Vegetation Index (NDVI), has substantially increased in the Sahel region between 1982 and 1999. Candidate explanations for the greening trend include a return to higher rainfall conditions, production stimulation through increased atmospheric CO2, and land use change caused by agricultural intensification or extensification, brought on by shifting social and economic conditions and mass migrations [Olsson et al., 2005]. Currently, these hypotheses remain untested.
 The positive trend in vegetation greenness may have implications for the global carbon cycle and greenhouse forcing. Tropical deforestation must lead to a large release of CO2 from ecosystems to the atmosphere, but this CO2 source is not apparent in atmospheric inverse model calculations [Schimel et al., 2001]. Rather, inverse modeling studies indicate a neutral overall behavior of the tropics with respect to carbon cycling during the 1980s and 1990s, suggesting the existence of a carbon sink somewhere in the tropics [Schimel et al., 2001]. J. W. Seaquist et al. (manuscript in preparation, 2005) suggested that a considerable fraction of this sink might be associated with the greening of the Sahel.
 We tested the extent to which driving forces acting directly on natural ecosystem processes in the Sahel can account for the observed greening trend and its interannual variability. We further tested the hypothesis that the Sahel has recently acted as a net sink for carbon in the tropics. This was accomplished by simulating vegetation dynamics and carbon exchange in the region over the period of the satellite data record with a dynamic global vegetation model, LPJ-DGVM [Smith et al., 2001; Sitch et al., 2003], using data for climate, soil texture, and atmospheric CO2 concentrations as the sole model inputs. In order to discriminate between different climatic factors as potential controls of the observed trend and variability, we performed simulations varying only one of the driving variables, while holding the other variables constant at 1982–1998 average values.
2. Satellite-Derived Greenness Data
 We used the Global Inventory Modeling and Mapping Studies (GIMMS) NDVI dataset derived from NOAA/AVHRR data [Tucker et al., 2005]. The data are 15-day maximum-value composites of 8 km resolution, and have been corrected for sensor differences, satellite sensor drift, and effects of major volcanic eruptions during the period. The processing scheme used empirical mode decomposition to identify and remove parts of the NDVI signal that are most related to the satellite drift [Pinzon, 2002]. We extracted peak NDVI values for the growing seasons 1982–1998 using the TIMESAT program [Jönsson and Eklundh, 2004], and aggregated these over the study area.
3. Vegetation Model
 LPJ-DGVM is a coupled biogeography-biogeochemistry model, which incorporates process-based representations of terrestrial vegetation dynamics and biogeochemical cycling. A full description of the model is given by Sitch et al. . Processes governing water uptake by vegetation were updated by Gerten et al. . The model is used by a broad community of global-change researchers and has, for example, been shown to successfully reproduce the interannual global exchange of CO2 between the biosphere and the atmosphere [Sitch et al., 2003] and the observed high-latitude vegetation greening trend in the 1980s' and 1990s' [Lucht et al., 2002]. A version adapted for regional-scale analyses, LPJ-GUESS [Smith et al., 2001], correctly predicts the magnitude and seasonal pattern of NEE at a range of flux measurement sites in temperate forests [Morales et al., 2005].
4. Modeling Protocol and Climate Data
 LPJ-DGVM was driven with monthly data for air temperature, precipitation, and sunshine hours from the global CRU05 dataset (0.5° resolution [New et al., 2000]). Monthly temperature, precipitation, and sunshine data were interpolated to provide quasi-daily time series. Annual global atmospheric CO2 concentrations were obtained courtesy of the Carbon Cycle Model Linkage Project [McGuire et al., 2001]. Climate data for a 1000-year model spin-up were derived from the same data by repeating detrended values of the climate input for 1901–1930, using pre-industrial CO2 content of the atmosphere.
 In order to assess the relative importance of different climatic variables, factorial experiments were performed in which the model was driven by changes in individual driving variables – temperatures, precipitation, sunshine or CO2 concentrations – while keeping remaining driving variables constant at 1982–1998 average values. Information on land use was not included; the model was applied so as to simulate natural ecosystem dynamics.
 The study was conducted for the Sub-Saharan region between 12°W and 35°E, and 10°and 16°N, spanning all of the arid and semi-arid Sahel. The southernmost, wetter part of the Sahel was excluded because, in this area, vegetation cover can be so high that the relationship between green biomass and NDVI saturates, making vegetation changes potentially difficult to discern [Field et al., 1995].
 The model generally reproduced the observed pattern of interannual variability and the overall trend in vegetation greenness between 1982 and 1998 (Figure 1). Averaged over the study area and period, peak-season NDVI was 0.41, and simulated peak-season LAI was 3.5. The only clear discrepancy between the satellite- and model-based time series occurred in 1989, when the model predicted a strong increase in vegetation greenness compared with the preceding year, while NDVI strongly decreased. This discrepancy could be the results of a delayed response of the model to a strong increase in mean precipitation (45%) between 1987 and 1988 (Figure 2): high precipitation in 1988 led to establishment of woody plant saplings in grassland ecosystems, with initially low leaf area index (LAI), followed by an increased LAI in 1989. The increased tree cover was not maintained in subsequent years as modeled tree density rapidly adapted to climate through growth-efficiency-related mortality. Alternative explanations are possible; for example, a build up of grass fuel over the dry period in the mid-1980s might have led to increased burning and an associated reduction in biomass not captured by the model. However, in the absence of data on wildfire frequency in the region, this hypothesis remains speculative. LPJ-DGVM simulates biomass consumption by wildfire, but the modeled values for consumed biomass accounted only for a minor fraction of total carbon fluxes and did not vary substantially between years.
 The simulations varying one climatic factor at a time revealed that the modeled overall vegetation changes were almost entirely attributable to changes in precipitation (Figure 3). This result is consistent with an increase in mean annual precipitation throughout the study period (Figure 2). Changes in temperature and sunshine did not cause any consistent trend in modeled greenness. Increasing atmospheric CO2 caused a slight linear increase in greenness (Figure 3).
 Integrated over the entire study period, the Sahel was predicted to have released 0.20 Gt of carbon to the atmosphere, but modeled net ecosystem carbon exchange (NEE) had a tendency to decrease, signifying an increased relative carbon sink over time (Figure 4). For the years 1990 to 1994, the model predicted an average carbon sink of 0.0032 Gt year−1, which is less than one percent of the sink that Schimel et al.  estimated for the same period for tropical and southern-temperate ecosystems globally.
 Our results strongly suggest that precipitation changes are sufficient to explain most of the change in vegetation greenness in the Sahel between 1982 and 1998, with CO2 having a minor positive effect. The close match between the independent time series from the satellite measurements and vegetation model simulations is evidence against a major influence of land use changes as a driver of changing vegetation greenness over the Sahel as a whole. Changes in human land use may, however, have a stronger impact on the vegetation in certain areas within the Sahel.
 Modeled NEE values indicate that the Sahel has most likely not contributed to the carbon sink in tropical ecosystems in the 1980s and 1990s. These model estimates of NEE cannot be validated directly because field measurements are not available for this region, and atmospheric transport models cannot resolve the Sahel [Schimel et al., 2001]. However, the fact that LPJ-DGVM fields for NEE have been successfully validated against observational data on both the site scale and aggregated for the whole globe (see section 3 above) suggests that the model generally predicts reasonable NEE values. Furthermore, the modeled increase in relative sink activity over time (Figure 4) is consistent with NEE estimates by Seaquist et al. , who applied a satellite data-driven light use efficiency model coupled with a model of soil carbon turnover to a smaller window within the Sahel.
 The present study complements the findings of Lucht et al. , who adopted a similar modeling approach using LPJ-DGVM to investigate greening of high-latitude ecosystems in the northern hemisphere. Lucht et al. showed that the high-latitude greening trend between 1982 and 1998, which is apparent in satellite observations [Myneni et al., 1997], was primarily driven by changes in temperature and growing season length. The results of the current study point, in contrast, to water availability as the primary agent of vegetation change in the Sahel. This region is particularly vulnerable to climatic change [Intergovernmental Panel on Climate Change, 2001]. Understanding the functioning of the ocean-atmosphere-biosphere system affecting the Sahel is therefore an important challenge for environmental sciences.
 We are grateful to Compton Tucker, NASA/GSFC, for granting us access to the 8-km GIMMS NDVI.