Fire and large mammal herbivores (LMH) are the principal top-down forces maintaining savanna structure.  Nonetheless, experiments designed to investigate interactions between fire and LMH are rare in savannas, and relationships between environmental variation and biodiversity in the context of fire and LMH are poorly understood.  This study addresses these gaps by manipulating the presence of LMH and early and late dry season fires in a tropical African savanna.  In addition, this work simultaneously explores environmental variables including soil and foliar quality, vegetation cover, and nearby water sources to more holistically describe factors affecting savanna functioning and biodiversity. 

After one year of experimental treatments, changes in vegetation were already apparent.  Shrub abundance and richness and grass richness were higher in the absence of LMH, while grass biomass increased three-fold in burned plots as compared to unburned plots.  Foliar nutrients tended to increase in open plots, while phenolics decreased.  Amphibian abundance decreased with early burns and was higher with LMH.  In contrast, small mammal abundance and richness increased without LMH and with time since fire.  Richness and foraging of LMH were highest after late burns.

These results demonstrate ecosystem-wide effects of LMH, illustrating the importance of considering multiple taxa when designing fire management programs. For example, burning negatively affected amphibians and small mammals and changed vegetation at the same time it increased LMH foraging.  In the long-term, this experiment will shed light on interacting effects of fire and LMH on savanna biodiversity and function.  

Study site

The Gorongosa Savanna Ecology Experiment is in a Rift Valley Lowland Woodland landscape (Lötter et al., 2021) in Gorongosa National Park (GNP), Mozambique (-18°58', 34°20', ~25 m a.s.l.).  The experimental area covers ~197-ha.  Woody plant communities include a mixed broadleaf savanna dominated by Combretum adenogonium (Combretaceae) which transitions to a more open savanna co-dominated by Acacia (Vachellia) robusta and Acacia (Senegalia) nigrescens (Fabaceae).  The most common shrub is Capparis erythrocarpos var. rosea (Capparaceae), and common grasses include Megathyrsus maximus, P. infestum, P. coloratum, Urochloa mosambicensis, and Digitaria eriantha.  Soils are silty clay loams.  The average annual temperature is 24.1° C, and average annual precipitation is 1053 mm.  The area is a mesic savanna with a dry season extending from April to November; 86% of rain falls between November and March.  Between 2001 and the start of the experiment in 2020, the study area burned every 2-3 years, and the last fire recorded before the experiment was in 2016 (Table S2).  The biomass of large mammal herbivores derived from full aerial counts of 129,393 hectares of savanna habitat in Gorongosa in 2020 and 2022 averaged 5,938 kg/km².  This biomass figure is moderate to high in comparison to areas with similar mean annual precipitation (East 1984, Hempson et al. 2015b).

Study design

In December 2020, three 30 x 30 m exclosure (fenced) plots were established in ten replicate blocks; fences were 2 m tall, and the top wires were electrified to prevent elephant damage.  Each block consisted of three exclosure plots and three open plots and averaged 1.15 ha in size.  Blocks were randomly located within the study area, maintaining a minimum distance of 100 m between them.  Plots within the blocks were a minimum of 20 m apart.  They were located to avoid large termite mounds and to contain at least one tree.  Treatment levels were randomly assigned to the plots.  One pair of open and closed plots was subject to a cold burn (beginning of the dry season) on 18-20 May 2021, and another pair of plots was subject to a hot burn (middle of the dry season) on 15-16 September 2021.  The first rain after the cold burn was 19 mm on 22-23 May 2021.  After the hot burn, the first rain was 5.3 mm on 19 September 2021.  Two control burn plots, one fenced and one open, burned due to escaped embers during the hot burn, so they were treated as hot burn plots in analyses.

Soil and vegetation measurements

Soil samples were taken before the experiment and one year after the first cold burns for C and N analyses.  In March 2021, before the exclosures began to take effect and before any experimental burns, three soil samples were collected from 2-15 cm depth at random locations within three randomly selected plots per block.  In May 2022, three soil samples were collected from all plots.  Samples were mixed at the plot level, dried in an oven at 50 °C, sieved, and sent to the Central Analytical Facilities at the University of Stellenbosch for C and N analyses following standard protocols using Vario EL Cube (Elementar Analysensysteme GmbH, Germany) elemental analysis.

Shrub richness and abundance were measured in February-March of 2021, just after the exclosures were installed, and again in February-May 2022.  Woody species classified as shrubs were perennial plants that consistently flower below 1.3 m and often do not have a clearly dominant stem.  Abundant species in our dataset were Capparis erythrocarpos, species of Grewia, Salvadora persica, and Dichrostachys cinerea.  Shrubs were measured in four 5 x 5 m subplots located 5 m from the edge of each corner of the plots.  Diameter at breast height (DBH) of all trees in the plots was measured in December 2020, before the exclosures were installed, and in December 2021, one year after the exclosures were installed.  Data were used to calculate the size corrected relative growth rate (RGR) and the change in basal diameter.  RGR was calculated by subtracting initial diameters from final diameters and dividing the response by the initial diameter.  Basal area measurements included all individuals present in 2020 and 2021 (including new recruits in 2021).  Basal area was summed at the plot level for analyses, and the average RGR per plot was used in analyses.  The area of all Capparis erythrocarpos var. rosea individuals in the plots was estimated by measuring the length and width of the plants at the same time DBH measurements were made.

Grass biomass and percent cover of grasses and herbaceous plants (forbs) were measured in February-March 2021.  These data were used for all pre-burn values through May 2021.  Biomass data were collected again in June, July, September, October, and December 2021, and between February and April 2022 (each plot was measured once in this last period).  Grass and herbaceous plant cover were measured again in June, August, September, October, and December 2021 and again between February and April 2022 (each plot was measured once in this period).  Biomass data from July represented July and August in analyses of fauna (see below).  Percent cover data from August were also used for both July and August.  Grass biomass, grass cover, and forb cover data representing November in analyses of fauna were measured in October (see below).  To measure grass biomass, grass height was measured with a disc pasture meter every 4 m along seven parallel transects spaced 4 m apart for a total of 56 points per plot.  The average height per plot was calculated and converted to dry weight of grass biomass based on a calibration curve developed with samples collected in the experimental area (Trollope and Potgieter, 1986).  Grass and forb cover were measured in 16 1 m² sub-subplots within each plot.  A sub-subplot was located in each corner of the 5 x 5 m subplots in which shrubs were measured (Appendix 1: Figure S1).  Cover was measured by visually estimating the percent of the quadrat occupied by grass or forbs.  One to two weeks after each fire, the area burned was measured in the same sub-subplots as grass cover.  The height of the burn scars on all trees was also measured.

Leaf samples were collected for elemental analysis (C, N, Ca, K, Mg, P, Si, and Na) and for analysis of total phenolics.  Five focal species representing the most abundant grasses and woody species were selected for analyses, Megathyrsus maximus (Poaceae), Urochloa mosambicensis (Poaceae), Capparis erythrocarpos var. rosea (Capparaceae), Acacia nigrescens (Fabaceae), and Acacia robusta (Fabaceae).  Undamaged leaves were randomly collected from all plots in which the plants were present in February 2022.  Samples from trees were collected from the lowest branches.  Samples were dried at 40 ̊C.  Leaf samples were ground to a fine powder in coffee grinders and were sent to the Central Analytical Facilities at the University of Stellenbosch for analyses.  Percent C and N were quantified using a Vario EL Cube Elemental Analyzer (Elementar Analysensysteme GmbH, Germany).  The concentrations (mg/kg) of other elements were measured with inductively coupled plasma mass spectroscopy using an Agilent 7900 inductively coupled plasma mass spectrometer (Agilent Technologies, Inc, USA) following the lab’s standard procedures.  Total phenolics were quantified using the Prussian blue assay (Price and Butler, 1977).  Extractions followed Hattas et al. (2005), and absorbance was determined using a spectrophotometer set at a wavelength of 720 nm (Shimadzu UV-1800).  Concentrations of total phenolics were determined using a calibration curve based on Gallic acid (Hattas and Julkunen-Tiitto 2012).

Insect sampling

Crickets (Gryllidae:  Orthoptera) were the focal insect group in this study because they are reliable indicators of habitat quality (Andersen et al., 2001) and are omnivorous and polyphagous, and are therefore less affected by the local plant assemblage in a given plot (Capinera, 2020).  In addition, crickets are relatively easy to sample with standardized, passive methods (Clayton, 2002).  Gryllidae is species-rich and widely distributed across Africa (Robillard et al., 2014; Jaiswara et al., 2018), with 1083 species described from the continent (Cigliano et al., 2023). 

Ground-dwelling crickets were collected during the dry season from April to November 2021 (with the exception of July) and after the rains began in January 2022.  Crickets were collected in nine pitfall traps per plot separated from each other by 5 m along two diagonal lines across the plots.  Traps consisted of 8 cm diameter cups inserted 13 cm into the soil and filled with water and detergent.  Two collections (April and May) were conducted before the cold burn.  The first post-cold burn collection was done one week after the fires (the last week of May).  Three weeks after the cold burn, another collection was done, and collections continued monthly until the hot burn in September.  Insects were collected two weeks after the hot burn, and collections continued monthly through November.  The pre-cold burn collections were each two days, and all subsequent collections were seven days.  The abundance and richness of insects in analyses was therefore corrected by the number of trapping days.  All insects were identified to morphospecies and are stored in the E.O. Wilson Biodiversity Laboratory in GNP.

Herpetofauna sampling

Amphibians and reptiles were assessed with visual searches and pitfall traps monthly.  Visual searches were performed nine times for amphibians, in March, May, June, July, August, September, October, and November 2021 and in February 2022.  Reptile data were collected in the same months with the exception of March 2021.  The same pitfall traps used in cricket sampling (described above) were used for herpetofauna sampling.  Visual searches consisted of searching each plot for 20 minutes.  All potential microhabitats (e.g., trees, leaves, shrubs, leaf litter, dead tree trunks, holes in the soil, seasonal pans) were searched.  All frogs, lizards, and snakes were recorded and photographed when possible.  Searches took place in the morning from 7:00 to 11:00 and in the afternoon from 15:00 to 18:00.  Five plots were searched each day, three in the morning and two in the afternoon; each plot was searched once in the morning and once in the afternoon every month. 

Habitat characteristics that may influence the presence of amphibians and reptiles were also measured.  In the same sub-subplots where grass and forb cover were measured, leaf-litter cover and litter depth were measured.  All downed tree trunks in the plots were counted, and their diameters were measured 1.3 m from the base.  Pans (seasonal ponds that serve as potential amphibian breeding sites) within 200 m of the center of each plot were counted using QGIS v. 3.20, based on GNP’s map of water sources. 

Small mammal sampling

Small mammals were sampled between May and October 2021 using Sherman traps (23 x 9 x 8 cm) baited with rice and sesame seeds mixed with peanut butter.  Six traps were spaced 9 m apart along two 28 m transects separated by 9 m in each plot (three traps per transect).  The traps were left open for three consecutive nights in each sampling period.  During the first sampling period, traps were set in all the plots.  One week after the cold burn, traps were set in the burned and control burn (no fire) plots.  One week after the hot burn, only the plots subject to the late season fire and the control burn plots were sampled.  All small mammals captured were identified, marked with a permanent marker, and released 100 m from the point of capture.

Large mammal sampling

Large mammal herbivores were surveyed using camera traps with a manufacturer-reported detection distance of 25 m and a maximum sampling radius of 120 m (Vikeri E2 Trail Camera, USA) in the unfenced plots.  Cameras were mounted at a height of 1.5 m in the NE corner of each plot, facing SE and angled slightly downward.  When triggered, the cameras took three photos, with a minimum 30 s interval between trigger events.  The cameras were active for three sampling periods: 19 March - 24 April 2021 (before the cold burn), 8 June - 8 September 2021 (after the cold burn), and 29 September 2021 - 30 January 2022 (after the hot burn).  Large herbivore species richness was determined for each sampling period (including warthog, bushpig, red duiker, oribi, impala, bushbuck, nyala, waterbuck, sable, kudu, wildebeest, buffalo, and elephant).  For the four most abundant species (warthog, impala, nyala, and kudu), a measurement of foraging intensity was calculated based on the number of individual animals photographed in a foraging posture during the sampling period (animals were counted once per detection event but counted multiple times if they appeared in sequential detections and were still foraging).