Complexity in water and carbon dioxide fluxes following rain pulses in an African savanna

The idea that many processes in arid and semi-arid ecosystems are dormant until activated by a pulse of rainfall, and then decay from a maximum rate as the soil dries, is widely used as a conceptual and mathematical model, but has rarely been evaluated with data. This paper examines soil water, evapotranspiration (ET), and net ecosystem CO(2) exchange measured for 5 years at an eddy covariance tower sited in an Acacia–Combretum savanna near Skukuza in the Kruger National Park, South Africa. The analysis characterizes ecosystem flux responses to discrete rain events and evaluates the skill of increasingly complex “pulse models”. Rainfall pulses exert strong control over ecosystem-scale water and CO(2) fluxes at this site, but the simplest pulse models do a poor job of characterizing the dynamics of the response. Successful models need to include the time lag between the wetting event and the process peak, which differ for evaporation, photosynthesis and respiration. Adding further complexity, the time lag depends on the prior duration and degree of water stress. ET response is well characterized by a linear function of potential ET and a logistic function of profile-total soil water content, with remaining seasonal variation correlating with vegetation phenological dynamics (leaf area). A 1- to 3-day lag to maximal ET following wetting is a source of hysteresis in the ET response to soil water. Respiration responds to wetting within days, while photosynthesis takes a week or longer to reach its peak if the rainfall was preceded by a long dry spell. Both processes exhibit nonlinear functional responses that vary seasonally. We conclude that a more mechanistic approach than simple pulse modeling is needed to represent daily ecosystem C processes in semiarid savannas. ELECTRONIC SUPPLEMENTARY MATERIAL: The online version of this article (doi:10.1007/s00442-009-1405-y) contains supplementary material, which is available to authorized users.