In order to calculate the share of observed CO₂ variability controlled by terrestrial ecosystem physiology we used state-of-the-art estimates of carbon flux anomalies from oceans, land-use change, anthropogenic emissions, and fires and reduced the measured atmospheric CO₂ growth rate anomalies accordingly (see figure 1). An amount of -0.91 GtC/yr of the observed anomaly of -1.61 GtC/yr is found to have been caused by changes in NPP and Rh. The LPJ model computes -0.86 GtC/yr (fig. 2): it is in quantitative agreement with the observations. Higher than normal oceanic carbon uptake and reduced anthropogenic emissions contributed to the anomalous carbon flux in addition to vegetation productivity and soil respiration (fig. 1). There are no indications that wild fires contributed significantly to the observed post-Pinatubo flux anomalies 14.

What was the cause for this large anomalous sink – changes in NPP or Rh? The model shows that the pronounced reduction of atmospheric CO₂ increase results from reduced soil respiration whereas NPP did not change (cf. 151617).

The physiological response of the land biosphere during the strong 1997/98 El Niño event was diametrically opposite to that of the 1992/93 period. A large anomalous flux of 2.82 GtC into the atmosphere was measured in 1998 of which 1.62 GtC can be assigned to vegetation growth and soil decomposition processes of the terrestrial biosphere (fig. 1). LPJ simulates a value of 1.85 GtC due to stronger soil respiration that was partly counteracted by increased photosynthetic activity (fig. 2). The remaining 1.2 GtC of the anomalous land source consists of a small contribution from anthropogenic emissions (about 0.2 GtC) while the rest is attributable to carbon emissions from extensive fires in tropical, subtropical and boreal regions 418(fig. 1).

The quality of modelled soil carbon decomposition is less certain than that of NPP. Studies show, however, that the temporal variability of modelled vegetation activity is in good agreement with independent satellite data 1719, indicating that the temporal variations of the associated simulated NPP are likely reliable. Because measured and modelled NEE, i.e. the difference of Rh and NPP, are found to be very similar in magnitude, we suggest that our simulation of the magnitude of anomalous soil respiration is plausible.

While the results show that soil processes controlled the increased land sink it remains unclear why NPP did not react to the climatic anomalies and what the relative importance of changes in temperature and precipitation were. We carried out a factorial experiment where either post-Pinatubo temperature or precipitation anomalies (that is, only the 1992–94 anomalies) were removed in the simulations, i.e. monthly values of temperature and precipitation were replaced by their corresponding 1979–2003 averages. These simulations help to clarify how relevant the changes in temperature and precipitation were for the carbon budgets of of vegetation and soils. We find that if the global cooling after the Pinatubo eruption is removed (NEE T const.) the terrestrial carbon sink declines together with Rh, and higher values of NPP are achieved (fig. 3). When the observed precipitation anomalies, associated with a weak El Niño event that occurred during the same time period, are eliminated, negative temperature anomalies alone trigger a much larger sink than observed (fig. 3, P const). The increased net flux of carbon from the atmosphere into the biosphere results from enhanced NPP whereas Rh remains unchanged. In summary, it appears that the anomalously large sink post-Pinatubo sink was caused mainly by the dampening effects of lower temperatures on soil microbial activity (fig. 2), but that post-Pinatubo water-limitation of NPP weakened the strength of the sink.

Warmer and wetter than usual conditions during 1998 stimulated both NPP and Rh in the wake of the large El Niño, with a particularly pronounced effect on soil respiration (fig. 2). The small anomalous increase in precipitation in the climate data deviates from the general assumption that El Niño episodes in the Amazon basin are characterized by extended dry seasons, lower wet season precipitation and shifts in spatial rainfall distribution 3. The strong sensitivity of heterotrophic respiration to soil moisture content has also been observed in eddy covariance measurements 20. Due to teleconnections El Niño periods also affect climatic conditions in the extratropics. For example, vegetation activity in the northern temperate regions is positively correlated with ENSO during the northern summer 21. A pattern of negative NPP anomalies in the tropics and positive anomalies in the temperate zones is also produced by the LPJ model. However, the simulated global carbon balance is mainly determined by the differential response of tropical NPP and Rh. In the northern mid-latitudes, higher CO₂ uptake by vegetation is counterbalanced by increased respiration due to higher temperatures (fig. 4). The dominant contribution of tropical ecosystems to the large land-atmosphere flux during El Niño conditions due to the opposite variation of NPP and Rh has been noted in other modelling studies as well 1322.

Removing precipitation anomalies from the climate data results in a small terrestrial carbon sink that is triggered by increasing NPP in the tropics while reducing the carbon flux from soils (fig. 4, P const). The influence of the temperature anomalies is more important outside the low latitudes. Eliminating the positive temperature anomaly of 1998 reduces Rh significantly in the mid-latitudes whereas the carbon fluxes in and out of tropical ecosystems become balanced (fig. 4, T const). Thus the 1998 carbon source resulted from two mechanisms, a limitation of vegetation activity and a stimulation of soil decomposition in the tropics as well as temperature-driven acceleration (possibly via teleconnections) of heterotrophic respiration relative to NPP in northern temperate regions.

The LPJ Dynamic Global Vegetation Model 910 is a biogeochemical model of fluxes of carbon and water in terrestrial vegetation and soils. Carbon uptake during photosynthesis is estimated using the Farquhar-Collatz scheme which is coupled to two soil layers 2425. Assimilated carbon is allocated to four pools (leaves, sapwood, heartwood and fine roots) following allometric and functional relationships 9. Carbon from dead biomass enters above- and belowground litter pools and is then transferred to a fast and slow decomposing soil carbon pool. Soil organic matter decomposition is calculated using a modified Arrhenius formulation 26 which implies a decline in apparent Q₁₀ with temperature, as well as an empirical soil moisture relationship 27.

We performed simulations with 0.5 degrees spatial resolution (59199 grid cells) using an interpolated climatology of monthly values of temperature, precipitation and radiation 2829. A land cover data set produced at the University of Maryland provided a realistic distribution of global crop lands 30. The LPJ-DGVM has been extensively validated using various data from atmospheric measurements, active and passive remote sensing data, and flux measurements. It was shown that the model is capable of simulating large-scale structure, distribution and phenology of global vegetation 917 as well as the inferred seasonal cycles of soil moisture 31, evapotranspiration and runoff 10.