Elephants and Fire as Causes of Multiple Stable States in the Serengeti-Mara Woodlands

Abstract

(1) Multiple stable states in ecosystems have been proposed on theoretical grounds, and examples have been offered, but direct tests of the predictions are lacking. A boundary between states exists if: (i) a system when disturbed from one state to another does not return to its original state once the cause of the disturbance returns to its original value; and (ii) a second factor takes over and holds the system in the new state. We examine these predictions for two stable states in the woodlands of the Serengeti-Mara ecosystem in East Africa. (2) Woodlands in natural areas of savannah Africa have declined over the past 30 years. Three general hypotheses have been proposed: (i) expanding human populations have concentrated elephants into protected areas, elephants then caused the decline of woodlands but man-induced fires prevented regeneration (two stable states); (ii) fires caused the decline and also prevented recovery (one stable state); (iii) fires caused the decline while elephants inhibited recovery through density-dependent mortality of seedlings (two stable states). (3) Two time periods the 1960s when woodlands changed fastest and the 1980s when grasslands prevailed, produced for specific hypotheses. (i) ''The 1960s elephant hypothesis'' and (ii) ''the 1960s fire hypothesis'' hold that elephants and fire, respectively, caused woodland change. (iii) The ''1980s elephant hypothesis'' and ''the 1980s fire hypothesis'' hold that these factors, respectively, prevented woodland recovery. (4) From experiment and observation of seedling recruitment, mortality due to combinations of burning rates, elephant browsing, wildebeest trampling, and antelope browsing was estimated and used to model tree population dynamics; predictions for rates of decline and increase were compared with independent estimates from aerial photographs. (5) Maximum rates of elephant and antelope browsing could not have caused the observed decline of woodlands in the 1960s. The most conservative burning rates in the 1960s, without elephants, could have caused a decline consistent with the 1960s fire hypothesis. (6) The combined impact of fire and browsing most closely matched the observed rate of woodland loss. (7) Wildebeest grazing in the 1980s reduced dry grass and minimized fire incidence. the model predicted that fire mortality and wildebeest grazing could not maintain the present grassland state. (8) The present high elephant density was sufficient to prevent an increase in the woodlands consistent with the 1980s elephant hypothesis. Wildebeest trampling and other browsers ensures that the vegetation is currently stable in a grassland state. (9) Thus, an external perturbation, such as fire, was necessary to change the vegetation from woodland to grassland. Elephants were unable to cause such a change. Once the grassland was formed, however, elephants were able to hold it in that state. These results are consistent with the third general hypothesis that there are two stable states of woodland and grassland, the latter maintained by herbivores. (10) Simulation of conditions in the 1890s suggests that the rinderpest epidemic combined with elephant hunting could have caused the woodland regeneration observed before the 1950s. Therefore, (i) savannah woodlands may regenerate in pulses as even-aged stands, and (ii) there may have been more grassland in Africa before 1890. This longer time-scale view of the dynamics of vegetation has implications for the conservation of elephants and their habitats.