• C4 grass;
  • fire;
  • low CO2;
  • palaeoecology;
  • savanna


The distribution and abundance of trees can be strongly affected by disturbance such as fire. In mixed tree/grass ecosystems, recurrent grass-fuelled fires can strongly suppress tree saplings and therefore control tree dominance. We propose that changes in atmospheric [CO2] could influence tree cover in such metastable ecosystems by altering their postburn recovery rates relative to flammable herbaceous growth forms such as grasses. Slow sapling recovery rates at low [CO2] would favour the spread of grasses and a reduction of tree cover. To test the possible importance of [CO2]/fire interactions, we first used a Dynamic Global Vegetation Model (DGVM) to simulate biomass in grassy ecosystems in South Africa with and without fire. The results indicate that fire has a major effect under higher rainfall conditions suggesting an important role for fire/[CO2] interactions. We then used a demographic model of the effects of fire on mesic savanna trees to test the importance of grass/tree differences in postburn recovery rates. We adjusted grass and tree growth in the model according to the DGVM output of net primary production at different [CO2] relative to current conditions. The simulations predicted elimination of trees at [CO2] typical of the last glacial period (180 ppm) because tree growth rate is too slow (15 years) to grow to a fire-proof size of ca. 3 m. Simulated grass growth would produce an adequate fuel load for a burn in only 2 years. Simulations of preindustrial [CO2] (270 ppm) predict occurrence of trees but at low densities. The greatest increase in trees occurs from preindustrial to current [CO2] (360 ppm). The simulations are consistent with palaeo-records which indicate that trees disappeared from sites that are currently savannas in South Africa in the last glacial. Savanna trees reappeared in the Holocene. There has also been a large increase in trees over the last 50–100 years. We suggest that slow tree recovery after fire, rather than differential photosynthetic efficiencies in C3 and C4 plants, might have been the significant factor in the Late Tertiary spread of flammable grasslands under low [CO2] because open, high light environments would have been a prerequisite for the spread of C4 grasses. Our simulations suggest further that low [CO2] could have been a significant factor in the reduction of trees during glacial times, because of their slower regrowth after disturbance, with fire favouring the spread of grasses.