The problem of tree-grass coexistence in savannas has attracted many theoretical solutions. Unidimensional niche separation models can be made to fit local observations in almost all cases, but tend to fail once extrapolated. Savannas are so widespread and diverse that a multiple-constraint model is needed if generality is the objective
Results/Conclusions

The core constituents of a general savanna model are competition for a seasonally-limited water supply, and the differential response of trees and grass to fire. In addition, competitition for nitrogen and light and interactions between fire and herbivory are necessary elaborations. The competition between trees and grass for water, nitrogen and light is spatially heterogeneous and temporally pulsed. Simplistic models, with for instance a single vertical axis of separation, do not do justice to the observed complexity of this competition. The minimum axis set is three-dimensional: vertical (hig, low, shallow, deep); horizontal (beneath tree canopy, between canopies) and temporal (wet period, dry period). A probabilistic rather than deterministic modelling framework is preferred.  The above points are illustrated with examples.