In savanna ecosystems, fire and herbivory alter the competitive relationship between trees and grasses. Mechanistically, grazing herbivores favor trees by removing grass, which reduces tree-grass competition and limits fire. Conversely, browsing herbivores consume trees and limit their recovery from fire. Herbivore feeding decisions are in turn shaped by risk-resource trade-offs that potentially determine the spatial patterns of herbivory. Identifying the dominant mechanistic pathways by which fire and herbivores control tree cover remains challenging, but is essential for understanding savanna dynamics. We used an experiment in the Serengeti ecosystem and a simple simulation driven by experimental results to address two main aims: (1) determine the importance of direct and indirect effects of grass, fire and herbivory on seedling establishment; and (2) establish whether predators determine the spatial pattern of successful seedling establishment via effects on mesoherbivore distribution. We transplanted tree seedlings into plots with a factorial combination of grass and herbivores (present/absent) across a lion kill-risk gradient in the Serengeti, burning half of the plots near the end of the experiment. Ungrazed grass limited tree seedling survival directly via competition, indirectly via fire, and by slowing seedling growth, which drove higher seedling mortality during fires. These effects restricted seedling establishment to below 18% and, in conjunction with browsing, resulted in seedling establishment dropping below 5%. In the absence of browsing and fire, grazing drove a 7.5-fold increase in seedling establishment. Lion predation risk had no observable impact on herbivore effects on seedling establishment. The severe negative effects of grass on seedling mortality suggests that regional patterns of tree cover and fire may overestimate the role of fire in limiting tree cover, with regular fires representing a proxy for the competitive effects of grass.

Data were collected within the Serengeti National Park (hereafter “Serengeti”), Tanzania, between March 2018 and December 2019. Our study site consisted of a subset of the ~206 camera traps comprising the Snapshot Serengeti (SS) grid. We used kill risk probability density estimates for each camera in the SS grid extracted from a kernel density distribution created using over 2500 historic lion kill locations of 6 common herbivore species. We selected 80 cameras in this section of the grid and categorized lion kill risk as high-, medium- or low-risk based on kill risk probability density functions. We then used stratified random sampling to select 48 final cameras (16 high-risk, 16 medium-risk, 16 low-risk) within a ~460 km2 area as our focal experimental sites. The grass community in the study area is dominated by Themeda triandra, Pennisetum mezzianum, Panicum maximum, Eustachys paspaloides and Digitaria macroblephara, and has burned approximately every 3 years since 2000.

Tree-seedling survival data

We focused on the seedling-sapling recruitment phase of tree demographics. We set up a fully factorial grass competition (grass vs. no grass) × herbivore exclusion (present vs. absent) experiment within each site, resulting in four 1m x 1m plots per site. In the grass competition treatment, grass was either completely removed or present, and in the herbivore exclusion treatment herbivores were either excluded using cages or allowed to feed freely. At the onset of the long rains, when trees normally recruit from seed (February 2018), we transplanted two V. robusta and two V. tortillis seedlings within each plot (15 cm from each corner), giving a total of 768 seedlings overall (2 species x 2 seedlings per plot x 4 plots per site × 48 sites). We had previously grown the seedlings under ambient light conditions and watered them to soil saturation every two days for 4 weeks prior to transplanting.

To non-destructively estimate aboveground grass biomass in each plot, we used an indirect approach based on light attenuation through the grass canopy. We sampled seedling height, basal diameter and grass biomass of each plot using the same protocol every three months until July 2019, at which time half of the sites were burned. Although efforts were made to limit wildfires, nine sites (four high-risk, four medium-risk, one low-risk) burned in unplanned fires during July 2018 (before we could apply the fire treatment). We binned remaining sites by predation risk category and used a random number generator to select five high-risk, five medium-risk, and eight low-risk sites that were burned in October 2019. We resampled plots one final time in December 2019 after two months of consistent rainfall when surviving burned seedlings had had sufficient time to resprout.

This data is avalaible in two .csv files:

serengeti.transplant.seedling.survival.csv - contains raw data of all seedlings through the entire study

fire.survival.csv - contains a subset of seedlings that were burned with pre and post burn survival recorded

Herbivore data

At each site we had one camera trap facing our experimental plots and we collected and processed all images over the study period. We used camera-trap occupancy data from these images for all mammalian herbivores larger than a dik-dik (Madoqua kirkii) and generated daily metabolic-adjusted herbivore indices (kg.d-1). We divided herbivores into mesobrowser, mesograzer, megagrazer and megabrowser guilds, but excluded giraffe from analysis as they are unlikely to feed on tree seedlings. We further split mesoherbivores into migratory or resident designations and calculated total daily metabolic-adjusted herbivore indices for each guild.

These data are available in the .csv file herbivore.mmass.csv

All the necessary information for these data is included in the ReadMe file.