Inferred controls on tree/grass composition in a savanna ecosystem: Combining 16-year normalized difference vegetation index data with a dynamic soil moisture model

[1] A two-layer soil moisture model was developed to infer the controls on the tree/grass coexistence in a savanna ecosystem. The model was applied along a mean annual precipitation gradient in southern Africa, known as the Kalahari Transect (KT), which follows a north-south decline in mean annual rainfall from 1600 mm/yr to 250 mm/yr between the latitudes 12 –26 S. Satellite-derived fractional covers for trees, grass, and bare soil were used as input to the model, with the fractional grass cover responsive to interannual variability in rainfall, as defined by observed statistics. Other inputs to the model included satellite-based radiation budget measurements and interpolated, groundbased rainfall measurements. The soil moisture model structure allows both trees and grass to have access to the upper zone, while trees alone can extract water from the lower zone. We focus our analysis on the wet season months of November–March, in which 87% of the annual rainfall is received along the KT. Simulations were performed on daily time steps for 16 years, representing a range of interannual rainfall variability, at 196 equally spaced positions along the transect. The results indicate that two distinct areas of the KT exist in terms of vegetation-rainfall relationships. North of 18 S, the trees and grass are rarely water stressed during the wet season and a large portion of the water balance is accounted for by leakage through the bottom of the root zone in sandy soils. The vegetation cover in the northern end of the transect does not reach its potential in terms of water exploitation and is most likely nutrient limited. In the southern portion of the KT, tree fractional cover is such that trees become water stressed in drier than average wet seasons and suffer no water stress during wetter than average wet seasons. The extent of the fractional grass cover for individual years is controlled by the water demand stress. We find that observed tree/grass fractional cover in the waterlimited region of the transect is best explained by considering, on an individual basis, an allowable range of stress for trees and grass, while minimizing the amount of water that is unexploited by the vegetation and lost as leakage from the root zone.