Study area and site selection
We conducted a plant litter decomposition study inside and outside the Serengeti National Park, Tanzania (latitude 2°40′ to 2°50′S and longitude 34°00′ to 34°90′E) during peak wet and dry seasons in 2017. We focused our study in two rainfall regions, a mesic region in the south-east and a wet region in the south-west of the Serengeti ecosystem (Table 1). Annual rainfall in these regions varies from 700 mm in the mesic east to 1300 mm in the wetter west (2015–2017 (Huffman 2017)). Rainfall varies seasonally with the majority of rainfall, up to 800 mm, occurring during the wet season between November and May. The dry season between June and October receives around 150 mm of rainfall, although the dry season is drier and longer in the mesic region compared to the wet region (Norton-Griffiths et al. 1975). Soil temperatures across the system are more uniform with small fluctuations around 29 °C in the upper 10 cm of the soil profile across the seasons (McNally et al. 2017). The elevation of our mesic and wet regions ranges between 1200 and 1670 m. Underlying soil types are primarily calcareous and stony leptosols with pockets of clay-rich and organic vertisols in the mesic region and organic planosols, sandy aeronsols and vertisols in the wet region (ISRIC 2018) (Table 1). Two-thirds of the Serengeti National Park is open wooded savannah, dominated by leguminous trees with nitrogen fixing symbionts such as Vachellia spp. and non-leguminous trees such as Commiphora spp. that are interspersed amongst a C4 grass dominated herbaceous layer.
Table 1 Characteristics of the seven study sites across agricultural, pasture and wildlife protected area land-uses at the borders of the Serengeti National Park, Tanzania. Study sites were split between two contrasting rainfall regions, wet and mesic region, with the addition of an intermediary mesic-wet region as a common garden in central Serengeti. Annual rainfall was estimated from satellite imagery of cloud cover between 2015 and 2017 (Huffman 2017). Year of last fire and fire frequency over a 16 year period was derived from MODIS MCD 45A burn product over the years 2000–2016. Livestock numbers are district-level averages from Tanzania National Bureau of Statistics 2012 census. Termite mound density (Odontotermes sp. and Macrotermes sp.) was estimated by counting mounds within a 50 × 50 meter square at all sites. Underlying soil type follow the World Reference Base (2006) Groups (ISRIC 2018). All other soil properties have been determined from soil samples collected for use in the common garden experiment. All soil properties are shown as mean ± 1 standard deviation Knowledge of the distribution and abundance of macrodetritivore species remains limited for the Serengeti ecosystem. There have been a handful of studies on macrodetritivores in the Serengeti mainly investigating litter and detritus decomposition by termite species (i.e. Macrotermes sp. and Odontotermes sp.) or on the diets of insectivorous mammals, for example aardwolf consumption of Trinervitermes spp. termites (de Visser et al. 2015; Freymann et al. 2008; Freymann et al. 2007; Freymann et al. 2010; Kruuk and Sands 1972; Smith et al. 2019). Inside wildlife protected areas, the majority of wild herbivores (including elephants, buffalos, impalas and various species of antelope) occur at low densities, with the exception of migratory wildebeest and zebra (Hopcraft et al. 2015; Sinclair et al. 2007). Within wildlife protected areas, landscape-scale spatial patterns of litter removal by macrodetritivores overlap with wild herbivore movements (de Visser et al. 2015; Freymann et al. 2010). Wildlife protected areas are regularly managed through burning, but fire management is presumed to have limited impact on rates of litter removal by macrodetritivores (Davies et al. 2013). The dominant land-use outside wildlife protected areas is agropastoral, comprising mosaics of small-scale agricultural holdings intermixed with livestock pastures (Veldhuis et al. 2019) (Table 1). Annual aboveground grass biomass production can be similar on pasturelands (470 g m−2 yr−1) and wildlife protected areas (515 g m−2 yr−1); however, livestock consume on average 70% of aboveground biomass in pasturelands compared to 40% by wildlife in protected areas (Arneberg et al. unpublished data). Intense livestock rearing removes stubble from agricultural land and reduces fuel loads on pasturelands and results in infrequent or no fires on agropastoral land (Veldhuis et al. 2019).
Within the mesic and wet rainfall regions, sites were selected across three human land-uses: agricultural land, pastureland and wildlife protected areas. Within each rainfall region all land-uses were within 10 km of each other to minimize spatiotemporal variation in rainfall between land-uses. Termites were found in the soil at all sites suggesting presence of termite-driven litter decomposition processes. All sites had Macrotermes and/or Odontotermes termite mounds, although we did not quantify the termite activity within mounds (Table 1). Agricultural sites were primarily used for growing maize (Zea mays) with minimal intercropping, but occasionally with beans and vegetables. Agricultural land was managed by hand hoeing with limited use of tractors and without pesticides or manmade fertilizers. The influence of illegal livestock grazing on sites inside wildlife protected areas was minimised by selecting sites a minimum of 9 km into wildlife protected areas in the mesic region, and in close proximity to a ranger post in the wet region.
We selected four replicate sites within each land-use (agricultural land, pastureland and wildlife protected area), each comprising an area of approximately 50 m2. All sites were a minimum of 500 m apart from one another, except for the agricultural sites. For these, we selected two agricultural fields in each rainfall region which were managed by the landowner of the adjacent pastural site. Each agricultural field was divided in two, thus creating paired agricultural sites on each agricultural field approximately 100 m apart. In total we had 24 sites (2 rainfall regions × 3 land-uses × 4 replicate sites).
Experimental design
To investigate the influence of season, rainfall region and land-use on plant litter decomposition we buried litterbags during both the wet season (from late January to early March 2017) and the dry season (from July to September 2017). Within each site, plots for litterbag burial were selected following a random cardinal direction and number of paces from the centre of the site. All plots were a minimum of 1 m apart and 2 m away from the nearest termite mound or tree canopy edge, with 100 m as the furthest distance from mound/tree. For the wet season there were eight plots per site and for the dry season seven plots per site.
We used a modified version of the globally standardized Tea Bag Index (Keuskamp et al. 2013). The Tea Bag Index uses two types of tea litter with distinct qualities: (1) green tea (Camellia sinesis; EAN no.: 8722700055525 Lipton® Tea) with high cellulose content (46.8% carbon (C), 4.1% nitrogen (N) and 11.5 C:N ratio, particle size ~6 mm2) and expected fast decomposition, hereafter referred to as ‘labile litter’; and (2) rooibos tea (Aspalanthus linearis; EAN no.: 722700188438 Lipton® Tea) with high lignin content (48% C, 1.2% N and 39.2 C:N ratio, particle size ~3 mm2) and expected slow decomposition, hereafter referred to as ‘recalcitrant litter’ (Keuskamp et al. 2013). The Tea Bag Index has based the lability and recalcitrance of tea litter types on rates of decomposition by soil microbes, while litter preferences of detritivores may differ from microbes. Several macrodetritivores, e.g. millipedes, termites and woodlice, prefer recalcitrant litter types (Hättenschwiler and Gasser 2005; Peguero et al. 2019; Sitters et al. 2014). For example, termite species ferment litter within the gut through a mutualistic relationship with microorganisms (i.e. bacteria, archaea and fungi) (Abe et al. 2000). Due to this fermentation process, termites avoid substrates high in readily digestible sugars (Abe et al. 2000), yet other macrodetritivore species can differ in litter preferences (Hättenschwiler and Gasser 2005).
Litterbag material was made of woven nylon with a mesh size of 0.25 mm, allowing access to soil microbes, microdetritivores and roots, but not macrodetritivores. However, prior work in the Serengeti found that termites eat through nylon litterbags to access plant litter creating holes ~0.55 mm2 in size (Smith et al. 2019). To quantify the contributions of macrodetritivores to litter decomposition, we used a stainless steel metal mesh treatment with an aperture size of 0.3 mm designed to exclude the head-width of the smallest foraging termites (Smith et al. 2019; Teo et al. 2020). Our litterbag approach targeted termites, but could equally apply to other savannah macrodetritivores with strong mandibles such as beetle larvae, millipedes and woodlice. Within each 20 × 20 cm plot we buried four litterbags in a factorial design using two pairs of labile and recalcitrant tea litter, one pair open to macrodetritivores (hereafter ‘accessible to macrodetritivores’) and the other pair excluding macrodetritivores with metal mesh (hereafter ‘excluding macrodetritivores’) (sensu (Griffiths et al. 2019); Fig. 1). In total 1488 litterbags were buried as part of the main experiment.
Following the Tea Bag Index protocol for warm climates (Keuskamp et al. 2013), one litterbag was placed in each corner of the excavated plot at a soil depth of 8 cm and incubated for approximately 2 months (52 days in the wet season and 69 days in the dry season). The majority of leaf litter decomposes on the soil-surface in tropical ecosystems. Burial of litter avoids decomposition by superterranean detritivore species, fire and UV degradation (Austin 2011; Cornwell et al. 2009; Davies et al. 2013). Nevertheless, we opted to follow the Tea Bag Index methodology and assumed observed decomposition processes were likely to be similar to those acting on root litter decomposition in savannahs (Smith et al. 2019). Upon collection, litterbags were taken out of the metal mesh, brushed clean of any adhering soil, plant roots and termites, then placed in paper bags and air-dried (30–40 °C) within 4 days of collection.
Common garden experiment
To further disentangle climatic effects of season and rainfall region and edaphic effects of land-use in the main experiment, an additional common garden experiment was established. The common garden was located near to the Serengeti Wildlife Research station in Seronera in the central part of the Serengeti National Park (Fig. 1). This involved decomposing litter in soil transplanted from the different rainfall regions and land-uses to a single location. Transplanting soil allowed us to examine the direct climate response of litter decomposition by microbes and microdetritivores, controlling for soil properties, and the indirect response of macrodetritivores as soil fauna originating from the common garden could enter transplanted soil. From the common garden, we expected that if rainfall was the main driver of litter mass loss, decomposition by microbes and microdetritivores or macrodetritivores would be similar in the main experiment and common garden when rainfall was similar between the two experiments. Here, litter mass loss from the common garden plotted against litter mass from the main experiment would follow a one-to-one line. Conversely, if there was less rainfall in the main experiment than common garden, or vice versa, then litter mass loss would deviate from a one-to-one line. If soil properties, transplanted microbial community or the local macrodetritivore community were the main drivers of litter mass loss, we would expect similarities or deviations from the one-to-one line to be site-specific or relate to land-use rather than variation in rainfall between the two experiments. The common garden site comprised a total of 50 m2 with four experiment blocks (Fig. 1). Each block was approximately 2 m2 and located a minimum of 5 m apart from one another. Within each block we established seven plots, the same area and size as the main experiment plots though excavated slightly deeper to ensure that the litterbags decomposed in transplanted soil, thus totalling 28 plots.
In the common garden experiment, 24 out of 28 plots were randomly assigned to be filled with soil from sites in the main experiment. Approximately 25 l of fresh soil were collected down to a depth of 20 cm at each site in the main experiment. Excavated soil was transported in loosely sealed plastic buckets to the common garden within 5 days. The remaining four common garden plots, one plot per block, were re-filled with local soil to serve as controls (Fig. 1). In each plot, four litterbags – a combination of labile and recalcitrant tea litter, with and without metal mesh to exclude macrodetritivores – were buried in soil immediately after creating the plot. Litterbags were incubated for 2 months and collected following the protocols outlined above. The common garden experiment was repeated for both the wet and dry season, removing old soil and re-collecting fresh soil for each season. In total, 224 litterbags were buried in the common garden experiment.
Soil moisture and rainfall
At the start of incubation and at litterbag collection, spot measurements of soil moisture were taken using hand-held probes in every plot. Soil moisture was measured via electrical conductivity (±0.1%) at a depth of 5.5 cm (ML3, Delta-T, Cambridge, U.K.). All measurements were taken between 7:15 h and 18:30 h during daylight hours. One permanent logger was established in one site per land-use within each rainfall region, to measure soil moisture throughout the experiment. Soil moisture was measured via electrical conductivity using a Decagon Device Em5b Analog data logger and GS1 water content sensor with a probe length of 5 cm (± 0.03 m−3 m−3 equivalent to ±3% volumetric water content in mineral soils). To reduce the visibility of data loggers due to risk of theft, loggers were buried next to the base of trees. Extended buried cables were used so that the probe was placed outside of the tree canopy and between 1.5 and 2 m from the tree trunk. Loggers were regularly checked and replaced. During the litterbag incubation period we only had missing data for soil moisture for agricultural land in the mesic region and pastureland in the wet region during the dry season due to repeated logger theft. To obtain a comparable zero measure across loggers, all soil moisture logger readings were adjusted for differences in soil water holding capacity by subtracting the lowest recorded value from all values for each logger soil type.
Rainfall for the wet and dry season incubation periods were obtained from satellite-based daily rainfall estimates from NASA’s Goddard Earth Sciences Data and Information Services Centre (Huffman 2017), based upon half-hourly measurements of cloud cover retrieved using multi-satellite microwave data at 10 × 10 km resolution. Previous work at the study sites showed significant positive correlation between these remote satellite based estimates of rainfall and soil moisture content (Smith et al. 2020). From daily satellite rainfall estimates we calculated cumulative rainfall for each seasonal incubation period averaged at the site-scale.
Measurements
Litter decomposition was calculated as ash-corrected percentage mass loss for the duration of each incubation period. Prior to burial, all litterbags were weighed (±0.001 g) with tea litter weights calculated by deducting the standard weight of nylon mesh, cord and label −0.25 g (Keuskamp et al. 2013). After the decomposition experiment, litterbags were oven-dried at 60 °C for 48 h and re-weighed. Litter was then extracted from the litterbag and weighed separately. Due to termite and other macrodetritivore intrusions into litterbags, remaining plant litter needed to be corrected for the weight of soil debris. Decomposed litter and debris inside litterbags was homogenized by pestle and mortar. Subsamples of homogenized litter were burned in a furnace at 550 °C for 4 h to determine Loss of Ignition (LOI). The remaining inorganic mineral ash was used to correct for amount of soil in the litterbags. Litter from 10 undecomposed labile and recalcitrant litterbags were also combusted via LOI to estimate undecomposed litter ash content. Litter mass differences of ash-corrected undecomposed and decomposed litter were used to calculate ash-corrected percentage litter mass loss.
Soil texture, carbon and nitrogen concentrations were determined from soil samples collected for the common garden experiment (sampling outlined above). Soil was sieved to 2 mm to remove stones and homogenised using pestle and mortar. Soil texture was determined using wet season samples only, following the pipette method (Gee and Bauder 1986). In brief, deionized water and hydrogen peroxide were added to 10 g of soil, which was heated until the organic material was fully oxidized. Water was added to each sample, rather than hydrochloric acid due to high pH, and the resultant solution went through a sedimentation analysis by repeatedly removing solution and heating to determine percentage of clay, silt and sand by weighing the precipitate. Soil carbon and nitrogen concentrations were determined for both wet and dry season soil samples. Soil subsamples of 16–22 mg were analysed for carbon and nitrogen concentrations by dry-combustion using an automated elemental analyser (vario MICRO cube, Langenselbold, Germany).
Statistical analyses
Tea litter from 1604 litterbags (out of 1712 buried) were recovered from the main decomposition experiment and the common garden experiment across both seasons. The effects of land-use, macrodetritivore exclusion, season and rainfall region on litter mass loss were analysed separately for labile and recalcitrant litter types using generalized linear mixed models. Percentage litter mass loss was fitted using a Beta distribution transforming mass loss to values between 0 and 1, thus ensuring model predictions were bounded between 0 and 100% after back-transformation. In our full models, fixed effect terms included: land-use (agricultural, pastureland and wildlife protected area), macrodetritivore exclusion (litter accessible to or excluding macrodetritivores), season (wet and dry) and rainfall region (wet and mesic) along with two and three-way interactions. Inclusion of rainfall region as a fixed term captured part of the spatial design of the experiment, and the remaining spatial structure was incorporated into random components, namely burial plot nested within replicate site. Soil moisture spot measurements were analysed separately using a generalized linear mixed model fitted with a Gaussian distribution and the same model structure outlined above with the omission of macrodetritivore exclusion, but with the inclusion of soil sand content and soil carbon-to-nitrogen ratio.
The common garden experiment was analysed by subtracting the mass loss in the main decomposition experiment from the common garden experiment. Mass loss in the main experiment was averaged at the site-scale because each site in the main experiment corresponded to a single plot in the common garden experiment. Difference in mass loss between the experiments were analysed using a linear-mixed model fitted with a Gaussian distribution. In the model season, rainfall region, land-use, and macrodetritivore exclusion were fixed effect factors without any interactions due to lower number of data points at the site-scale. Nevertheless, site was retained as random factor to account for paired litterbags accessible to and excluding macrodetritivores.
Final models were simplified following Akaike’s Information Criterion (AIC), removing terms from the full model to improve the model likelihood and lower AIC value. Fixed variables were retained if significant in Likelihood Ratio Tests. For the final model, significance of each term was assessed contrasting models using Maximum Likelihood with and without fixed factors to generate P-values (Bolker et al. 2009; Zuur et al. 2009). Significant differences within terms and interactions were obtained through multiple contrasts as a function of least square means. All analyses were carried out in R version 3.5.3 (R Foundation for Statistical Computing, 2019) with GLMM and LMM models tested using the ‘glmmTMB’ and ‘lmer’ functions in lme4 (Bates et al. 2015) and glmmTMB (Brooks et al. 2017) packages and model contrasts using the emmeans package (Lenth 2016).