Detecting process from snapshot pattern: lessons from tree spacing in the southern Kalahari

The spatial distribution of plants is often thought to be an indicator of u nderlying biotic and abiotic processes. However, there are relatively feu e xamples of spatial patterns being analysed to detect an underlying ecologic al process. Using the spacing of savanna trees in the southern Kalahari as an example. we applied methods of computer simulation modelling and point p attern analysis in an evaluation of their potential for identifying relevan t pattern generating processes from snapshot pattern. We compared real tree patterns from the southern Kalahari, derived from aerial photographs, with patterns produced from computer simulation experiments in an investigation of the following questions: does the present pattern of tree distributions allow us to characterize (1) the relative importance of the major driving forces (e.g., competition for moisture, grass fire, herbivory), (2) the spa tial dimensions and structures of the underlying processes, and (3) the act ual dynamic status of the ecological system (a phase of decline, increase o r constancy with respect to tree abundance)? The simulation experiments are based on a well established, spatially expli cit. grid-based model that simulates the vegetation dynamics of the major l ife forms under a realistic rainfall scenario of the southern Kalahari and under the impact of grass fires. herbivory and the formation of localized c lumps with increased tree seed availability. For a realistic range of param eters the simulation modal produces long-term coexistence of trees and gras ses with tree densities that correspond with densities observed in the fiel d. Both real tree distributions derived from aerial photographs and tree pa ttern produced bq the model are characterized by a tendency towards even sp acing at small scales, clumping at intermediate scales and randomness or cl umping at large scales. However. increasing the spatio-temporal correlation in the formation of seed patches in the model caused an increase in the te ndency towards clumping in the tree distribution whereas an increase in see d patch numbers led to a decrease in clumping. Within single simulation run s the tree pattern could change in response to the variable rainfall sequen ces and the corresponding differences in grass fire frequency: periods of s lightly increasing tree numbers caused bq higher precipitation were charact erized by an increase in tree clumping whereas periods of slightly decreasi ng tree numbers showed a tendency towards random or even tree spacing. Simu lating the transition of an open savanna to a savanna woodland show ed that the tree pattern in the transitional phase can be diagnostic of the underl ying process: If the transition was caused by improved moisture conditions the transitional phase was characterized bq increased clumping in the tree pattern. In contrast, a transition caused by an increase in the number of l ocalized tree wed patches led to a characteristic even spacing of trees. Even though the simulated savanna clearly showed non-equilibrium dynamics, simulation results indicate that the tree population in the stimulated area of the southern Kalahari is in a state of long-term tree-grass coexistence with the persisting structure of an open savanna system.