Summary 1Relationships between above-ground net primary productivity (ANPP) of grasslands and annual precipitation are often weak at the site level, with much of the inter-annual variation in ANPP left unexplained. A potential reason for this is that the distribution of precipitation within a growing season affects productivity in addition to the total amount. 2 We analysed long-term ANPP data for three southern African temperate grasslands (mean annual precipitation ranging from 538 mm to 798 mm) to determine the effects of precipitation event size, number and spacing relative to seasonal totals. 3 Ungrazed, non-manipulated treatments at each site showed contrasting results despite sharing a common, dominant species. At the driest site, a model combining average event size and number of events per growing season provided a substantially better fit to the ANPP data than precipitation amount (seasonal total). At the wettest site, the interval between events was the most important precipitation variable. Precipitation distribution was not important at the intermediate site where amount was the best predictor of ANPP. A limit to the size of precipitation events efficiently utilized for ANPP was evident for the driest site only. 4 At each site, experimental treatments that altered species composition and soil fertility had little effect on precipitation-ANPP relationships. The lack of consistency in the relative importance of the precipitation variables among sites suggests that local, edaphic factors modify precipitation-ANPP relationships. 5 This analysis demonstrates that the distribution and size of precipitation events can affect ANPP independent of precipitation amount. As altered precipitation regimes are forecast by global climate models, the sensitivity of ecosystems to precipitation distribution should be considered when predicting responses to climate change. 6 While mean values of precipitation, and other ecosystem drivers, are typically used to predict function at the level of whole ecosystems, our results show that more complex measures of environmental variability may be required to understand ecosystem function, and to increase the accuracy of predictions of ecosystem responses to global change.

Tree biomass is an integrated measure of net growth and is critical for understanding, monitoring, and modeling ecosystem functions. Despite the importance of accurately measuring tree biomass, several fundamental barriers preclude direct measurement at large spatial scales, including the facts that trees must be felled to be weighed and that even modestly sized trees are challenging to maneuver once felled. Allometric methods allow for estimation of tree mass using structural characteristics, such as trunk diameter. Savanna trees present additional challenges, including limited available allometry and a prevalence of multiple stems per individual. Here we collected airborne lidar data over a semiarid savanna adjacent to the Kruger National Park, South Africa, and then harvested and weighed woody plant biomass at the plot scale to provide a standard against which field and airborne estimation methods could be compared. For an existing airborne lidar method, we found that half of the total error was due to averaging canopy height at the plot scale. This error was eliminated by instead measuring maximum height and crown area of individual trees from lidar data using an object-based method to identify individual tree crowns and estimate their biomass. The best object-based model approached the accuracy of field allometry at both the tree and plot levels, and it more than doubled the accuracy compared to existing airborne methods (17% vs. 44% deviation from harvested biomass). Allometric error accounted for less than one-third of the total residual error in airborne biomass estimates at the plot scale when using allometry with low bias. Airborne methods also gave more accurate predictions at the plot level than did field methods based on diameter-only allometry. These results provide a novel comparison of field and airborne biomass estimates using harvested plots and advance the role of lidar remote sensing in savanna ecosystems.

Grazing represents the most extensive use of land worldwide. Yet its impacts on ecosystem services remain uncertain because pervasive interactions between grazing pressure, climate, soil properties, and biodiversity may occur but have never been addressed simultaneously. Using a standardized survey at 98 sites across six continents, we show that interactions between grazing pressure, climate, soil, and biodiversity are critical to explain the delivery of fundamental ecosystem services across drylands worldwide. Increasing grazing pressure reduced ecosystem service delivery in warmer and species-poor drylands, whereas positive effects of grazing were observed in colder and species-rich areas. Considering interactions between grazing and local abiotic and biotic factors is key for understanding the fate of dryland ecosystems under climate change and increasing human pressure.

No abstract