Me. Hochberg et al., INFLUENCES OF TREE BIOLOGY AND FIRE IN THE SPATIAL STRUCTURE OF THE WEST-AFRICAN SAVANNA, Journal of Ecology, 82(2), 1994, pp. 217-226
1 Using a spatially explicit cellular automaton model we explore the e
ffects of tree demography, fire-induced mortality, and seed dispersal
on the spatial spread of a single tree species in a humid savanna at L
amto in West Africa. 2 The model system is described by six parameters
and consists of a grass-surrounded square grid of connecting cells, e
ach being either inhabited by grass alone or by grass and an individua
l tree. In the baseline numerical simulations the tree can only recrui
t seedlings in immediately adjacent cells. These seedlings may perish
from annual grass fires in their first year of life if they are not pr
otected from the advancement of the fire by neighbouring reproductivel
y mature trees. 3 Based on preliminary parameter estimates from data c
ollected at field sites at Lamto, we predict that fire slows, but does
not stop, the spread of the tree. In the absence of fire the doubling
rate of the tree population is about 6 years, whereas we predict that
yearly fires prolong this to at least 30 years.4 The temporal dynamic
s of the tree population are fairly smooth and predictable as long as
there are more than c. 100 cells in the system. As the number of cells
is decreased below c. 100 the trajectories become increasingly variab
le from year to year. 5 Mortalities from fire act in an inverse spatia
lly density-dependent fashion, enhancing tree aggregation. The role of
fire in enhancing tree aggregation is supported by additional simulat
ions in which dispersal of seeds to non-adjacent cells can occur. When
a small amount of dispersal is possible the rate of tree population g
rowth is greatly accelerated as compared to when no such dispersal occ
urs. 6 We present several hypotheses to explain why the savanna at Lam
to is not tree-dominated as would be predicted by the model, discuss h
ow seed dispersal and fire influence tree dynamics, and make predictio
ns for future testing.