Journal of Ecology 2014, 102, 1171–1182

doi: 10.1111/1365-2745.12290

Functional attributes of savanna soils: contrasting effects of tree canopies and herbivores on bulk density, nutrients and moisture dynamics Ricardo M. Holdo1,2* and Michelle C. Mack3 1

Division of Biological Sciences, University of Missouri, 217 Tucker Hall, Columbia, MO 65211, USA; 2Animal, Plant, and Environmental Sciences, University of the Witwatersrand, Private Bag 3, Braamfontein, 2050 Gauteng, South Africa; and 3Department of Biology, University of Florida, 509 Carr Hall, PO Box 118525, Gainesville, FL 32611, USA

Summary 1. Savannas are highly heterogeneous tree-grass mixtures, and the structural variation imposed by a discontinuous canopy cover results in spatial variation in soil properties such as plant-available nutrients, temperature and soil moisture. Many savannas are also dominated by large vertebrate herbivores, which impose a different suite of effects on soil properties related to consumption, excretion and physical disturbance. 2. In nutrient-poor, water-limited systems, variation in soil resource availability may play a fundamental role in structuring plant communities, but the relative and interactive effects of key drivers of heterogeneity have received little attention in savanna ecosystems. 3. We investigated the independent and interactive effects of herbivores and tree canopy cover on multiple soil properties up to a depth of 10 cm, including bulk density, soil C pools, total and plantavailable nutrients, moisture dynamics and temperature at two contrasting sites in Kruger National Park, South Africa. We analysed samples from locations under and between adult tree canopies and inside and outside long-term herbivore exclosures using a nested design. 4. We found strong positive effects of tree canopies on total and labile pools of soil C and N and negative effects on plant-available P and soil temperature. Tree canopy effects on soil moisture availability were small but clearly detectable. During precipitation events, trees increased moisture inputs on granite soils but decreased moisture inputs on basalt soils compared to open sites. 5. Conversely, we found mixed evidence for herbivore effects on soil chemistry and temperature, but a clearer effect on soil moisture, with areas outside exclosures experiencing higher soil moisture retention, particularly immediately following precipitation events. Soil bulk density, a driver of both nutrient and soil moisture dynamics, was lower under trees and when herbivores were excluded. 6. Synthesis. Our results suggest that variation in tree cover is the dominant biotic driver of C, N and P dynamics in these savanna systems in the top 10 cm of soil, that herbivores primarily affect soil moisture content, and that canopy and herbivore effects tend to be additive rather than synergistic. Key-words: functional heterogeneity, grazers, Kruger National Park, microclimates, plant-available nutrients, plant–soil (below-ground) interactions, resource limitation, spatial heterogeneity

Introduction Savanna vegetation is characterized by a high degree of structural heterogeneity (Scholes & Archer 1997). This heterogeneity is primarily the result of small-scale variation in canopy cover, which in turn exerts control over biophysical processes (Sankaran, Ratnam & Hanan 2008; Brudvig & Asbjornsen 2009). It has been established, for example, that savanna tree *Correspondence author. E-mail: [email protected]

canopies diminish understorey solar radiation and temperature and enhance soil moisture availability and nutrient and carbon (C) concentrations, with consequences for herbaceous composition and diversity (Belsky et al. 1989; Vetaas 1992; Brudvig & Asbjornsen 2009). Many savannas, particularly in Africa, are also dominated by large herbivores, which impose their own effects on soils and plants, thus potentially enhancing or counteracting tree canopy effects (Vetaas 1992). When herbivores aggregate under trees, non-additive (i.e. synergistic or antagonistic) canopy–herbivore interactions may result

© 2014 The Authors. Journal of Ecology © 2014 British Ecological Society

1172 R. M. Holdo & M. C. Mack (Vetaas 1992). A key remaining challenge is to understand the relative contributions and interactions of tree canopies and herbivores to soil functional heterogeneity. Unlike grasslands or forests, which are treeless or have essentially closed canopies, respectively, savannas have a discontinuous tree layer over a ground layer of near-continuous grass cover (Scholes & Walker 1993). They are therefore highly patchy in terms of incoming radiation, precipitation (which may be intercepted or concentrated) and in the amount and quality of litter deposition both above- and below-ground. This spatial heterogeneity in the abiotic and biotic environment can in turn influence the underlying grassland matrix and soil (Scholes & Archer 1997). Tree canopy effects can influence soil nutrients and C via competition with grasses for resources (Knoop & Walker 1985; Riginos 2009), through variables that influence decomposition rates (such as soil moisture and temperature) and through leaching (Belsky et al. 1989, 1993). Trees can also redistribute nutrients and water through hydraulic lift (Schulze et al. 1998; Ludwig et al. 2003) and concentrate water flow via stem-flow (Vetaas 1992). Conversely, tree canopies intercept water, a substantial fraction of which may then be lost through evaporation prior to reaching the soil surface, particularly during small precipitation events (Rodriguez-Iturbe et al. 1999). Overall, subcanopy effects on soil nutrients have generally been found to be positive relative to inter-canopy sites (Belsky et al. 1989; Ludwig et al. 2004; Holdo, Mack & Arnold 2012), but effects on soil moisture have been less well described (Belsky et al. 1993; Belsky & Canham 1994). The role of herbivores on soil C and N dynamics has been quite widely studied in many grassland systems (Parton et al. 1987; Frank et al. 1994; Biondini & Manske 1996; Johnson & Matchett 2001). Herbivores have been shown to impose considerable effects on soil nutrients in savanna ecosystems when aggregated at high densities, such as in hotspots (McNaughton 1988; McNaughton, Banyikwa & McNaughton 1997; Augustine, McNaughton & Frank 2003) or near waterholes (Naiman et al. 2003), but their overall effects at large scales or at the scale of tree canopies and inter-canopy spaces are often not the focus of investigation. When examined across a range of systems, herbivores can exert either positive or negative (and sometimes neutral) effects on soil N (Milchunas & Lauenroth 1993). Their effects on soil P have been less well-studied, but these may be as or more important than effects on soil N, given the evidence for P limitation or colimitation of primary productivity in tropical savannas (Penning de Vries, Krul & van Keulen 1980; Chapin & McNaughton 1989; Craine, Morrow & Stock 2008). Herbivores may also potentially influence soil moisture dynamics by causing changes in plant transpiration as a result of grazing, and by altering interception in the herbaceous or tree layer through biomass removal (Coughenour 1984; Frank & Groffman 1998). Trampling, a relatively understudied ecological process (Cumming & Cumming 2003), may also plausibly affect soil hydrological properties by reducing rates of drainage to deeper layers, but generally, the role of herbivores for soil moisture dynamics remains largely unstudied.

In addition to the independent effects of tree cover and herbivore biomass, synergistic interactions between these two variables may occur through a variety of mechanisms. For example, canopy trees attract birds and other animals that may act as vectors for importing nutrients from the surrounding matrix (Vetaas 1992; Dean, Milton & Jeltsch 1999), and may influence the spatial distribution of foraging behaviour of shade-seeking herbivores (Vetaas 1992; Treydte et al. 2010). Shade-seeking herbivores may concentrate nutrients under canopies (Vetaas 1992) or otherwise magnify any baseline effects they may exert on soil moisture dynamics, thus exacerbating any pre-existing canopy effects. By and large, these interactions have not been quantified because of the lack of data for these two drivers at single sites. Much of the literature on tropical savannas has focused on the direct effects of disturbance and herbivores on vegetation structure (Hoffmann 1999; Archibald et al. 2005; Bond 2005; Bond, Woodward & Midgley 2005; Holdo 2007; Staver et al. 2009). In ecosystems with highly seasonal precipitation, however, soil moisture availability may be the key limiting resource for plant growth (Sankaran, Ratnam & Hanan 2008; February et al. 2013). Terrestrial ecosystems in general are also nutrient-limited (Vitousek & Howarth 1991), and nutrient to carbon ratios in soils play an important role in controlling nutrient availability (Parton et al. 1987). Spatiotemporal variation in nitrogen (N) in particular is widely associated with variation in plant community composition (Bobbink et al. 2010), and has been proposed to influence tree to grass ratios in savannas (Sankaran, Ratnam & Hanan 2008). Given the pervasive nature of resource limitation for vegetation, it is important to understand the factors influencing various resource pools in systems where such pools are likely to show a high degree of spatial heterogeneity. In savannas specifically, a key question is: how do some of the key drivers responsible for the heterogeneity that pervades this biome, such as herbivores and tree canopies, separately and jointly influence spatial variation in nutrient and C pools and soil moisture dynamics? In this study, we used ~35-year-old exclosures to explore the effects of herbivores, tree canopies and their interactions on nutrient and soil moisture dynamics on two soil types in the lowveld savanna of Kruger National Park (KNP), South Africa. Previous work in this system has examined canopy effects on soil nutrients and forage quality, either in isolation or in relation to fire (Coetsee, February & Bond 2008; Treydte, Grant & Jeltsch 2009; Holdo, Mack & Arnold 2012). A general finding is that trees are strongly associated with elevated C and N soil pools, and that fire effects are primarily indirect, and modulated by tree cover (Coetsee, February & Bond 2008; Coetsee, Bond & February 2010; Holdo, Mack & Arnold 2012). Here, we explored how herbivores and herbivore–tree interactions affect soil nutrient and C stocks and drive soil moisture dynamics. We asked two key questions: (i) How do herbivores affect total and plant-available nutrients and soil C pools, soil moisture availability and soil temperature? (ii) What is the relative importance of canopy versus herbivore effects on soil chemistry, moisture and temperature?

© 2014 The Authors. Journal of Ecology © 2014 British Ecological Society, Journal of Ecology, 102, 1171–1182

Herbivore and tree effects on savanna: soils 1173

Materials and methods STUDY SITES

We sampled soils and monitored soil temperature and soil moisture in and around two large long-term herbivore exclosures (Fig. 1) within the boundary of KNP. This park comprises 20 000 km2 of tropical to sub-tropical ‘lowveld’ (low-altitude) savanna along the South AfricaMozambique border in southern Africa. The climate is characterized by a well-defined summer rainy season (spanning the period from late October–May), with mean annual precipitation ranging between 10 m from the access road, >20 cm diameter at breast height; Fig. S1 in Supporting Information). We similarly identified and marked a matching tree outside the fence (O sites, with high herbivory). At each tree, we obtained three soil cores to a depth of 10 cm under the crown of the tree (the ‘under’ or U sample) plus 3 matching cores away from the tree, between the crowns of the target and neighbouring trees (the ‘between’ or B sample). We combined the U and B cores, stored the samples in airtight ziplock bags, and kept them refrigerated prior to transportation to the laboratory in the United States. We collected a fixed, known volume of soil with each core (262 cm3) and weighed the fresh composite sample prior to subsampling and reweighing between 100 and 200 g of soil for shipment and analysis. At the basalt exclosure, we restricted sampling to portions of the fence that had not been burned in several years to avoid confounding effects of fire (the exclosures are divided into blocks and burned in a rotation). Also, the effect of fencing out herbivores is quite dramatic at this exclosure (Asner et al. 2009), as evidenced by an almost complete lack of trees >20 cm d.b.h. outside the fence. In this case, we took the O site as the nearest tree (regardless of size) across the fence from the matching I site. Given that these trees were heavily-browsed and effectively lacked crowns of appreciable sizes, we only collected B samples in most cases. Strictly speaking, the herbivory effect is unreplicated in this study (given that we only sampled one exclosure per soil type), as landscape-level studies often are (e.g. Asner et al. 2009; Levick et al. 2009). In practice, however, the large size of the exclosures (>6 km of fencing, >250 ha) should allow pairs of sites (inside and outside

(b)

Fig. 1. Sampling locations at the herbivore exclosure sites: (a) Hlangwine (granite) and (b) N’washitshumbe (basalt). Labels correspond to trees around which paired under- (U) and between-crown (B) soil samples were taken. Circles identify trees with paired soil moisture and temperature sensors located in U and B locations. Trees with sensors were paired in sites across the fence boundary. © 2014 The Authors. Journal of Ecology © 2014 British Ecological Society, Journal of Ecology, 102, 1171–1182

1174 R. M. Holdo & M. C. Mack the fence) to be considered as independent samples for practical purposes, provided that they are located far enough from each other to minimize unwanted spatial autocorrelation and to capture a meaningful amount of the landscape heterogeneity present in the system as a whole. To help assess whether this was the case, we examined the scale of spatial autocorrelation across sampling sites for each exclosure by applying Moran’s I (Legendre & Legendre 1998) to the first principal component calculated across five soil variables (total C and N, labile C and N, and Mehlich-P). We found no significant spatial autocorrelation at any of the lag distances tested, including those corresponding to adjacent sites (~200 m). The trend in the pattern of autocorrelation also tended to alternate between being positive and negative as a function of increasing lag distance, which indicates that the exclosures were large enough to capture variation in soil properties across entire catenas, which are dominant mesoscale topographic properties of the landscape, further suggesting that the spatial scope of the enclosures was sufficient to encompass much of the spatial heterogeneity present in this system (Fig. S1).

SOIL MOISTURE AND TEMPERATURE

We collected soil moisture and temperature data at a subset of eight trees (four each inside and out) per exclosure (Fig. 1) over the rainy season, spanning ~200 days between November 2009 and June 2010. We selected trees using a stratified randomization protocol (one individual per group of four consecutive trees inside the fence, paired with its matching tree on the other side of the fence). At each tree, we buried two HOBO soil moisture probes (ECH2O EC-10; Decagon Devices, Pullman, WA, USA) vertically in the top 10 cm of soil, one under the tree canopy (the U probe) and one in the nearby intercanopy area (the B probe). Both probes were connected to a single HOBO Micro Station datalogger (Onset Computer Corporation, Bourne, MA, USA) attached to the tree trunk. At each exclosure, we connected a tipping bucket-type rain gauge to one of the dataloggers located inside the fence. For temperature measurements, we buried ibutton temperature dataloggers (DS1921G; Maxim Integrated, San Jose, CA, USA) at a depth of 5 cm next to (but not touching) each soil moisture probe. Soil moisture and rainfall data were logged at 5-min intervals, while temperature data were logged every hour. These sampling schemes allowed us to obtain ~205 days of soil moisture data (November–June) and ~85 days of temperature data (November– February) over the course of the growing season.

LABORATORY ANALYSES

In the laboratory, we passed soil samples through a 2-mm sieve and then subsampled them for determination of gravimetric water content (GWC) (by drying at 105 °C for 48 h) and total C and N. We used our water content data to convert the field-measured soil mass to dry mass. We calculated the bulk density for every sample by dividing the known volume of soil collected by its dry mass. Prior to this, we verified that the soil subsample shipped from the field had not lost moisture prior to weighing and drying in the lab. We ground air-dried subsamples with a mixer-mill (Spex Certiprep, Metuchen, NJ, USA), and analysed the ground soil with a Costech ECS 4010 Elemental Analyser (Valencia, CA, USA) at the University of Florida. We determined the mineralizable N fraction in soils (as an index of plantavailable N) from laboratory incubations. We analysed initial (in " field-moist soils) and final pools of inorganic N (NHþ 4 and NO3 ) on soil samples extracted with 50 mL of 2 M KCl. We moistened final (incubated) samples to an estimated 60% water holding capacity

(determined from a composite soil sample) with a fixed amount of deionized water (Ruess & Seagle 1994). We shook solutions for 2 h and left them overnight before filtering and freezing prior to analysis. We determined inorganic N concentrations in the extracts colorimetrically on an Astoria-Pacific autoanalyser (Astoria-Pacific, Clackamas, OR, USA). We also determined labile C pool size from the incubations following a method described by Lavoie et al. (2011). We analysed soil-respired CO2 on a LI-COR 6252 CO2 analyser (Lincoln, NE, USA) fitted with a calibrated injection port. We used extractable P from the Mehlich-1 procedure as an index of plant-available soil P (Kuo 1996). We analysed extracts colorimetrically with a microplate spectrophotometer (PowerWave XS Microplate Reader; Bio-Tek Instruments, Winooski, VT, USA).

STATISTICAL ANALYSIS

We used linear mixed models implemented in the R (R Development Core Team 2011) package nlme (Pinheiro & Bates 2000) to test for effects of herbivores (HERB, as captured by fencing) and canopy cover (CAN) on soil bulk density, nutrient concentrations and mineralization rates, and temperatures. We note that it would be more accurate to call the herbivore effect a fencing or removal effect, but find that the former description makes our results easier to follow and interpret. We included two temperature (mean daily temperature Tmean and daily temperature range Trange) and seven soil variables: bulk density, total C and N, in vitro mineralizable N (Nmin) and C (Cmin), Mehlich-P, and C:N. We conducted a first set of analyses with N, P and C data expressed on a mass basis (in g kg"1 of dry soil). The initial analysis showed a strong effect of both fencing and canopies on bulk density, at least on granite, so we repeated the analyses an area basis (g m"2 for the top 10 cm of soil) for all variables except C:N, which as a ratio is invariant with respect to mass or area. We logtransformed most soil variables prior to analysis, except for Nmin, which exhibited negative values in several cases (i.e. indicating immobilization), as well as Cmin and the soil temperature variables. We treated fencing and canopy position as fixed effects, and modelled two random effects: an individual tree effect (with paired U and B sampling locations) nested within a site effect (with paired I and O trees), and a site effect. For soil nutrients at the granite exclosure, we tested main effects and interactions among fixed effects. At the basalt exclosure, the near-absence of U samples outside the fence resulted in a highly-unbalanced design and demanded a two-part analysis: a test of canopy effects on I trees, and a test of fencing effects on B sites. Each of these analyses included a single random effect (individual trees in the first case and sites in the second). For the soil temperature data, we initially proposed to leverage the large number of temperature records in our data set into a repeated measures analysis, but an initial analysis showed high temporal autocorrelation in the model residuals over long (>20 days) time lags, even after detrending the data with periodic functions. When such autocorrelation patterns are observed over a few time lags, they can readily be accounted for by modelling the temporal dependency in lme (Pinheiro & Bates 2000). In this case, we opted for the more conservative approach of estimating mean values of Tmax and Trange per datalogger for the duration of the sampling interval, and treating these means as observation units in a linear mixed model. As with the soil nutrient data, the lack of U sites outside the fence on basalt required a two-part analysis of crown and fence effects on separate subsets of the data. After fitting the models, we visually and quantitatively examined plots of residuals versus fixed effects and quantile-quantile normal plots of both residuals and random effects to evaluate departures from normality and

© 2014 The Authors. Journal of Ecology © 2014 British Ecological Society, Journal of Ecology, 102, 1171–1182

Herbivore and tree effects on savanna: soils 1175 variance homoscedasticity (Pinheiro & Bates 2000). For the few cases in which serious outliers were detected, we repeated the analyses without the outliers to evaluate the robustness of the analysis to the presence of such values. In addition to testing for fencing and canopy effects, we tested for effects of soil organic matter quality on Nmin, to examine the extent to which any observed variation in N mineralization can be explained by litter quality effects on decomposition rates. We used two measures of soil organic matter quality: the log of the C:N ratio and the ratio of Cmin to C. We interpreted the latter as the ratio of labile or fast to recalcitrant soil C. We ran regression models using lme with the same random effect structure described above for canopy and herbivore exclusion effects.

MODELLING SOIL MOISTURE DYNAMICS

For the analysis of soil moisture dynamics, we initially proposed to follow the approach we adopted for the analysis of the temperature data. This was complicated by two factors. First, unlike temperature, soil moisture is subjected to episodic forcing by rainfall, which leads to a pattern of sudden spikes and decays in moisture content that appear to interact nonlinearly with the fixed effects of interest (e.g. canopy cover). This means that it is difficult or misleading to simply subtract the average soil moisture trend from each sensor and relate the resulting offset value to some independent variable. The problem is exemplified in Fig. S2, where two sensors (a B and a U sensor for the same tree) appear to exhibit quite different temporal patterns, with decay trajectories that cross over time (indicating different rates of decline). The second complication is introduced by gaps in our data, given that temporary failures of some of the soil moisture probes and the rain gauge at one of our sites resulted in significant periods without data for some sensors. Elimination of all days without a complete data set would have substantially reduced our power to make meaningful inferences. As a result of these challenges, we opted for a modelling approach that allowed us to combine a simple, well-established soil moisture dynamics process model (Rodriguez-Iturbe et al. 1999; Laio et al. 2001; Guswa, Celia & Rodriguez-Iturbe 2002) into a Bayesian statistical framework that allowed us to impute probability distributions for missing data points, and thus leverage the entire data set in the analysis. Our process model essentially identifies three phases or processes in a soil moisture dynamics time series: water input, and ‘fast’ and ‘slow’ output rates (Fig. S2). Inputs are driven by precipitation events (Fig. 2a) and the factors that modify the size of these events, such as canopy interception or stem flow. Outputs are controlled by drainage to deeper soil layers (in our case to depths below 10 cm), evaporation and plant transpiration [or evapotranspiration (ET) when the latter two are considered jointly]. Following a precipitation event (which leads to a spike in soil moisture content), there is an initial, fairly rapid drop-off in moisture (the fast output), due mainly to drainage as soils return to their field capacity, followed by a slower draw-down (the slow output), driven mainly by ET. In our model, we incorporated the canopy and fencing effects as potential modifiers of these processes and rates. For example tree canopies can modify inputs by locally diminishing or enhancing incoming precipitation amounts, and/or affect the fast and slow output rates that dominate when soils are above and below field capacity, respectively (a priori we link these rates with drainage to deeper layers and ET, respectively, and revisit these assumptions in the Discussion). We used a state space framework for the analysis. State space models combine a process model (in this case, our soil moisture dynamics model) with an

observation or data model (our sensor data). The state space framework allowed us to model both observation error (i.e. measurement or sensor error) and process error. Process error allows for the fact that, even in the complete absence of observation error, soil moisture dynamics will exhibit stochastic variability associated with unmeasured factors (Clark & Bjornstad 2004; Holdo et al. 2009). For example, transpiration might vary seasonally as a result of phenology, leading to temporal drift in soil moisture decay rate, even when soil moisture is measured with great accuracy. State-space models are also ideal for dealing with the second challenge outlined above, i.e. missing data (Clark & Bjornstad 2004), because the model effectively estimates (with error) the underlying ‘true’ soil moisture time series that provides the best fit to the process and observation models, even for time periods that lack observations. This is because the values that occur before and after a gap in the data for a particular sensor, as well as values obtained for other sensors, provide information about the state of the sensor during the interval with missing data. The larger the gap, the greater the uncertainty around these missing values, and the lower the leverage of these imputed points for model inference (e.g. Fig. 2b). Finally, the highly hierarchical nature of our design (i.e. random effects of trees nested within sites, probes nested within trees and repeated measures at the probe level) made the problem amenable to a Bayesian framework, with posterior distributions derived through Markov Chain Monte Carlo (MCMC) methods, implemented in WINBUGS (McCarthy 2007). The soil moisture model, which is explained in detail in Appendix S1, in principle should allow us to tease apart the independent effects of fencing and tree crowns on the inputs and outputs outlined above, because each of these processes can, to varying degrees, be inferred from particular features of the trajectories of maximum and mean daily soil moisture increases and declines. For example, if two sensors start at the same baseline soil moisture, but one exhibits a higher spike following a rainfall event than the other (Fig. S2), we can infer that the sites associated with each sensor modify inputs (e.g. through interception) in different ways. We incorporated the effects of the covariates on the three processes that drive the model by expressing three model variables as linear equations containing the two covariates (Appendix S1). For example, fast outputs are controlled by the saturated conductivity rate (Rodriguez-Iturbe et al. 1999), which in our model, rather than being treated as a fixed parameter, is a simple function of CAN and HERB. We conducted the soil moisture analysis slightly differently for the granite and basalt sites. We performed a single analysis at the granite exclosure, with additive effects of CAN and HERB on each of the soil moisture dynamics processes. We did not test for interaction effects to avoid overburdening an already-complex model. On basalt, the almost complete lack of canopy cover outside the fence meant that we lacked outside U data. We therefore divided the soil moisture analysis into two parts: an analysis of canopy effects inside the exclosure, and an analysis of fencing effects between B samples. Also, to test the ability of our approach to infer the parameters of the underlying process model, we generated a simulated data set with tree canopy and herbivore effects on inputs and fast outputs (Appendix S1). Our Bayesian approach was able to recover the underlying model parameters (Table S7). To obtain soil moisture data for the Bayesian analysis, we first calibrated our HOBO sensors with soil samples of known moisture content from the granite exclosure (GWC measured in the lab). We used: (i) soil cores collected to a depth of 10 cm from sites immediately adjacent to each sensor in March 2010, and (ii) containers with soil of contrasting soil moisture content into which we inserted a pair

© 2014 The Authors. Journal of Ecology © 2014 British Ecological Society, Journal of Ecology, 102, 1171–1182

1176 R. M. Holdo & M. C. Mack

Daily rainfall (mm)

(a) 80

Observed Imputed

60 40

Results

20 0

0

50

100

150

200

Day Obs U Obs B True U True B

(b) 0.6 0.5 0.4

S

0.3 0.2 0.1 0.0 0

50

100

150

200

150

200

Day (c) 0.6 0.5 0.4 0.3

S

of soil moisture dynamics. We ignored random effects, and only included one stochastic effect, the process error. Each time series was driven by the actual rainfall time series from our data set. We then compared mean soil moisture availability (computed across the 100 runs) under and between canopy sites and inside and outside the fence.

0.2 0.1 0.0 0

50

100

Day Fig. 2. Growing season rainfall (a) and soil moisture data for two trees (b, c) at the granite (Hlangwine) exclosure. Details about rainfall imputation are provided in Appendix S1. of moisture probes. We used a linear regression (R2 = 0.96) to transform raw values into GWC, multiplied by bulk density to get volumetric Water Content (VWC), and divided by soil porosity (which we estimated from the literature) to get our final metric, soil saturation S (Rodr!ıguez-Iturbe et al. 1999; Guswa, Celia & Rodriguez-Iturbe 2002). Given that bulk density varied as a function of CAN and HERB, we implemented the model both with site-specific bulk density values and with mean bulk density values. We used the latter case to test whether any canopy or fencing effects on soil moisture that we detected were driven purely by spatial variation in bulk density (which affects soil moisture) or not. The model is described in detail in Appendix S1. Given that we lacked a specific calibration for basalt, we treat our model results there as approximations, and emphasize relative contributions of the covariates rather than absolute soil moisture values generated by the model. To translate the various canopy and fencing effects into an overall, readily interpretable effect on moisture availability, we conducted a simple simulation for the granite exclosure in R, in which we took mean posterior values of the model coefficients and generated 100 time series

SOIL BULK DENSITY AND NUTRIENTS

Bulk density in the top 10 cm of soil was lower under canopies than between canopies at both exclosures, while herbivore presence increased bulk density on granite, but not on basalt. On a mass basis, we found no effects of herbivore exclusion (or herbivore by canopy interactions) for either total or labile pools of C and N or for Mehlich-P (Tables 1 and S1, S2). On an area basis, however, the effect of herbivore exclusion on bulk density modified the results on granite, with areas outside the fence (i.e. with higher herbivore biomass) having higher total C and N than areas inside. The canopy effects were consistent regardless of the basis used to express soil chemistry. Canopy effects were common on both soil types, with total and labile C and N tending to be higher under tree crowns (Tables 1 and S1, S2, Figs 3 and 4), and Mehlich- (extractable) P being higher between crowns on basalt (Tables 1 and S2, Fig. 4). Outlier removal only slightly altered the results; in particular, the removal of one prominent outlier in the Cmin analysis on basalt indicated that labile C may be lower outside the exclosure than inside on this soil type (F1,14 = 16.99, P < 0.005). The canopy effects on C and N were comparable in magnitude: C:N ratios did not vary with canopy position, or as a function of herbivore exclusion (Tables 1 and S1, S2). Nmin was unrelated to either C:N or Cmin:C ratios on either granite (C:N: F1,28 = 0.02, P = 0.9; Cmin:C: F1,28 = 0.48, P = 0.5) or basalt (C:N: F1,15 = 3.36, P = 0.09; Cmin:C: F1,15 = 0.29, P = 0.6), suggesting that canopy effects on N mineralization are not associated with variation in these aspects of soil organic matter quality as a function of canopy position. SOIL TEMPERATURE AND MOISTURE DYNAMICS

Tree canopies, but not herbivore exclusion, had clear cooling effects (~2 °C) on mean soil temperatures at both sites, and moderated temperature fluctuations at the granite (but not basalt) site (Tables S3 and S4). The state-space model estimated inputs and outputs (expressed as linear functions of CAN and HERB) for soil moisture availability over the course of the wet season. There was support on both granite and basalt for a negative effect of HERB on fast outputs (which we assume are driven by drainage to deeper layers) to a depth of 10 cm, as inferred from the mean value and credible intervals (which did not include 0) for the coefficient a2 (Table 2, Appendix S1). This negative effect means that herbivores reduce the fast output rate compared to a baseline rate given by an intercept a0 (Table 2). It also means that outside the exclosure, where

© 2014 The Authors. Journal of Ecology © 2014 British Ecological Society, Journal of Ecology, 102, 1171–1182

Herbivore and tree effects on savanna: soils 1177 Table 1. Effects of herbivore exclusion (HERB) and tree canopies (CAN) and their interactions on soil variables at two sites on contrasting soils (granite and basalt) in Kruger National Park, South Africa, analysed both on a per-mass and per-area basis Variable Basis

Soil type

Factor

Mass (g kg"1)

Granite

HERB CAN HERB 9 CAN HERB CAN HERB CAN HERB 9 CAN HERB CAN

Basalt Area (g m"2)

Granite

Basalt

BD

Total N

Total C

Mehlich-P

Nmin

O > I*** B > U**** ns ns B > U****

ns U> ns ns U> O> U> ns ns U>

ns U> ns ns U> O> U> ns ns U>

ns ns ns ns B > U* ns ns ns ns B > U*

ns U> ns ns U> ns U> ns ns U>

B***

B** I* B*

B**

B***

B** I* B**

B*

C:N

Cmin ns U> ns ns U> ns U> ns ns U>

B**

B** B*

B***

B****

B*** B****

B**

ns ns ns ns ns n/a n/a n/a n/a n/a

I, low herbivore biomass; O, high herbivore biomass; U, under canopy; B, between canopies. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

8.0

5.6

1.4 1.2

7.2 6.8

IB

OU

6.0

OB (e)

0.0 –0.4 –0.8 –1.2 –1.6 –2.0

IU

IB

OU

OB

IU

IB

OU

4.8 4.4

3.6

OB

IU

IB

OU

OB

IU

IB

OU

OB

(f)

2.0 1.2 0.4

–0.4 –1.2 –2.0

5.2

4.0

Cmin (g C m–2 d–1)

IU

0.4

–2.4

7.6

6.4

Nmin (g N m–2 d–1)

Ln Mehlich-P (g m–2)

(c) 6.0 Ln N (g m–2)

1.6

1.0

(d)

(b) 8.4 1.8

Ln C (g m–2)

Bulk density (g cm–3)

(a)

IU

IB

OU

OB

1.4 1.0 0.6 0.2

Fig. 3. Herbivore exclusion and tree canopy effects on soil variables at the Hlangwine exclosure (granite) in KNP: (a) bulk density, (b) total soil C, (c) total soil N, (d) Mehlich-P, mineralization rates for laboratory soil incubations of (e) N (Nmin) and (f) C (Cmin). Letters on the x-axis correspond to sites that are inside (I) and outside (O) the exclosure fence, and soils collected under (U) or between (B) tree canopies. Lines link samples (U and B) within individual trees and trees within sites (I and O). KNP, Kruger National Park.

herbivores occur at a higher population density, the top 10 cm of soil tended to remain wetter shortly after precipitation events compared to soils inside the fence. Conversely, we found a positive effect of CAN on fast output (compared to between-canopy areas) on granite (but not on basalt), suggesting that water flows through the upper soil layer faster under tree canopies than in the open (Table 2). We found no effects of any of the covariates on slow output at either site (Table 2). We found opposing effects of tree canopies on inputs on the two soil types: tree cover was associated with higher water inputs in the top 10 cm as a result of precipitation events on granite compared to open areas, but lower input rates on basalt. On granite, herbivore exclusion also affected input, but its effects opposed those of canopies: areas outside

the exclosure had lower inputs than those inside (Table 2), indicating that herbivores were associated with a reduction in the fraction of incoming precipitation making it into the top soil layer. This effect, while nonzero, was small, however; whereas the opposing effects of herbivores and canopies on fast output were comparable in absolute magnitude, the herbivore effect on input was small compared to the canopy effect (Table 2). All of the results reported so far were consistent regardless of the whether the soil moisture model used sitespecific or mean values of bulk density (Tables S5 and S6). The exception was provided by the fencing effect on input, which was predicted to be positive on basalt (i.e. opposite to what we found on granite) for the model with site-specific bulk density but unimportant when we used mean bulk density values (Tables S5 and S6). We note that our state-space

© 2014 The Authors. Journal of Ecology © 2014 British Ecological Society, Journal of Ecology, 102, 1171–1182

1178 R. M. Holdo & M. C. Mack

1.4 1.2 1.0 0.8

IU

IB

IB

(e)

2 1 0 –1 –2 –3

IU

IB

IB

OB

8.0

5.5

7.5 7.0 6.5

OB

3

(c) 6.0

Ln N (g m–2)

Ln C (g m–2)

1.6

8.5

IU

IB

IB

(f)

5 4 3 2 1 0 –1

IU

IB

IB

OB

5.0 4.5 4.0

OB

Cmin (g C m–2 d–1)

Ln Mehlich-P (g m–2)

(d)

(b)

Nmin (g N m–2 d–1)

Bulk density (g cm–3)

(a) 1.8

IU

IB

IB

OB

IU

IB

IB

OB

1.4 1.2 1.0 0.8 0.6 0.4 0.2

Fig. 4. Herbivore exclusion and tree canopy effects on soil variables at the N’washitshumbe (basalt) exclosure in KNP: (a) bulk density, (b) total soil C, (c) total soil N, (d) Mehlich-P, mineralization rates for laboratory soil incubations of (e) N (Nmin) and (f) C (Cmin). There were almost no sites outside the fence in the U category, so only under (U) versus between (B) canopy comparisons inside the fence (I) and inside versus outside (O) fence comparisons (between canopies) are depicted. Lines link samples (U and B) within individual trees for I sites and B sites (I and O) across the exclosure fence. KNP, Kruger National Park.

model contains fixed parameters (e.g. field capacity; Appendix S1) that would be expected to vary with bulk density, and we did not examine the effect of this dependence on our model inference. We note that these parameters mainly control when soil moisture output switches between the fast and slow phase, rather than the rate of decay itself (Appendix S1), so we would not expect that allowing these parameters to take sensor-specific values would have a qualitative effect on our conclusions. We note also that we tested for a correlation between mean GWC values per sensor and bulk density at the granite site, and found no relationship between the two variables for the subset of locations with both sensor and bulk density data (N = 16, R2 = 0.005). This suggests that our observed patterns are not simply spurious relationships driven by bulk density effects on sensor readings. The simulation results indicated that the mean net effects of tree canopies and herbivore exclusion on soil moisture over the course of wet season were small (mean = 4.8% higher at U vs. B sites and 11.6% higher at U vs. B sites). This suggests that the positive effects of canopies on water inputs at the granite site are largely cancelled out by their effects on the fast output rate, indicating that the main role of tree canopies is to accelerate water turnover through the top 10 cm of the soil rather than to affect overall moisture content.

Discussion SOIL NUTRIENTS AND BULK DENSITY

The finding that soil total C and N (as well as labile N, C and P) are affected by tree canopies reinforces previous findings from other studies as well as our own work on canopy–fire

interactions in KNP (Belsky et al. 1989; Ludwig et al. 2004; Treydte, Grant & Jeltsch 2009; Holdo, Mack & Arnold 2012). We propose that the observed patterns are primarily due to direct biotic effects of the trees, such as nutrient transport, or to tree–grass interactions. Tree effects on soil nutrient pools could also result from herbivore effects, effects of trees on soil organic matter quality, or be a consequence of microclimatic conditions under tree crowns, such as lower temperatures and higher soil moisture, or combinations of the above (Vetaas 1992; Belsky 1994; Coetsee, Bond & February 2010). Our results are in conflict with some of these proposed processes. We inferred elevated water inputs at granite sites, but not on basalt, where subcanopy areas show similar patterns of elevated C and N pools. Second, despite cooler subcanopy temperatures and slightly lower soil moisture availability, the U sites had higher rates of C and N mineralization than open (B) sites. This suggests that the canopy effect on the nutrient cycle may not be driven through abiotic control of the mineralization process. This is because while mineralization rates increase with soil temperature and moisture (Parton et al. 1987), we found higher labile C and N concentrations where both temperature and soil moisture were lower. Also, we found no relationship between N mineralization and our two indices of soil organic matter quality, suggesting that labile N pools are not controlled by variation in the quality of litter inputs under and between canopies. The lack of herbivore by canopy interactions in our results (to the extent that we were able to test for them) also suggests that canopy effects do not interact with any effects that herbivores may exert on soil nutrients, which in our study were largely absent. We found no evidence for any effects of long-term herbivore exclusion on labile pools of N or plant-available

© 2014 The Authors. Journal of Ecology © 2014 British Ecological Society, Journal of Ecology, 102, 1171–1182

Herbivore and tree effects on savanna: soils 1179 Table 2. Posterior distribution means and lower and upper 95% credible intervals (CI) for canopy (CAN) and herbivore exclusion (HERB) effects on three processes driving soil moisture dynamics Granite

Basalt CAN effects* CI

CI

Process

Factor

Par

Mean

2.5%

97.5%

Fast output

Intercept CAN HERB Intercept CAN HERB Intercept CAN HERB

a0 a1 a2 b0 b1 b2 c0 c1 c2

2.94 0.42 "0.48 "0.47 0.01 0.18 "1.24 0.27 "0.08

2.86 0.35 "0.56 "0.92 "0.27 "0.29 "1.31 0.19 "0.16

3.01 0.50 "0.40 "0.16 0.27 0.68 "1.17 0.35 0.00

Slow output

Input

Basalt HERB effects*

Mean 0.6 0.15 "25.6 "12.5 "0.45 "0.66

CI

2.5% 0.56 "0.38 "166.3 "154.0 "0.54 "0.78

97.5%

Mean

1.23 0.53 "0.3 75.8 "0.36 "0.55

2.5%

97.5%

0.78

0.50

1.03

"0.64 "89.5

"1.02 "235.0

"0.30 "3.7

"36.8 "0.43

"195.6 "0.51

114.7 "0.34

0.05

0.44

0.26†

Positive coefficients for CAN and HERB indicate positive effects of herbivores (O > I) and tree canopies (U > B) on a given process, respectively, compared to baseline values (given by intercepts). Coefficients shown in bold differ from zero. The results are for site-specific bulk density soil moisture conversion factors (see text for details). * CAN and HERB effects were examined separately on basalt due to the absence of U (under canopy) sites outside the fence at this site. † This nonzero effect does not hold when mean bulk density is used to calculate soil moisture from sensor data.

P. Herbivores are generally assumed to enhance plant-available N because the herbivore decomposition pathway shortcuts the plant litter pathway and returns N to the soil in inorganic form (urine) or as high-quality (i.e. low C:N) detritus (Frank et al. 1994; McNaughton, Banyikwa & McNaughton 1997; Johnson & Matchett 2001). Although herbivore effects on soil nutrients may be substantial when herbivores are aggregated, for example in hotspots (Augustine, McNaughton & Frank 2003; Naiman et al. 2003), it is less clear what their average effects are at landscape scales, and here we found no evidence for an herbivore effect on soil organic matter quality as measured by C:N ratios inside and outside exclosures. The majority of the Kruger landscape does not experience superabundant herbivore densities, and sites not close to water (such as our experimental exclosure sites) probably provide a more realistic assessment of the mean impact of herbivores. On basalt, we did find some weak evidence of a herbivore effect on labile C, but the effect was opposite to what we had predicted. We expected that higher nutrient turnover in areas with high herbivore population density might lead to larger pools of labile C and N. It is possible that what appears as a herbivore effect is actually a fire effect (consistent with an effect of fire on labile C on basalt, as reported in Holdo, Mack & Arnold (2012)). It has previously been shown that the basalt exclosure is associated with dramatic differences in tree cover on opposite sides of the fence (Asner et al. 2009; Holdo & Mack 2014). There may be a synergistic interaction with fire at work here, however. Loss of canopy cover outside the fence as a result of elephant activity favours grassland expansion, which can lead to hotter fires (with the caveat that the exclosures also exclude grazers, which tend to reduce fuel loads and may counteract the effects of browsers). These fires may accelerate tree cover loss and/or exert a differential impact on soils inside and outside the exclosure. Overall, we may draw an analogy with previous work on canopy–fire

interactions in this system, which suggested that fire effects on soil nutrients are mainly indirect and mediated through effects on canopy cover. At our basalt site, where browsing is intense, herbivores may drive similar changes in tree cover and cause canopy effects (which we studied at the scale of individual trees) to scale up at the landscape level. As a caveat to the lack of an herbivore association with nutrient pools, one of the most striking patterns in our study was the effect of both canopy cover and herbivore biomass on soil bulk density in the top 10 cm. The higher bulk density outside the fence may be associated with soil compaction as a result of trampling by herbivores. Support for this is provided by the contrasting relationship between bulk soil C and bulk density inside and outside the fence (Fig. S3). On granite (where the bulk density patterns were most prominent), we found a strong negative relationship between bulk density and soil C inside the fence, where herbivore biomass is very low, but this relationship disappeared outside the fence (Fig. S3), suggesting that C is the primary driver of bulk density variation when herbivores are absent, and that herbivores: (i) tend to disrupt this relationship and (ii) increase bulk density through soil compaction when they are present. Despite the lack of strong effects of the granite exclosure on woody vegetation structure, even decades after the exclosure was constructed (Asner et al. 2009), visible differences were apparent in terms of effects on the ground layer across the fence at this site, with clear evidence of trampling and abundant dung piles outside the exclosure. We note that although the canopy effects on bulk density were consistent across the two soil types, the herbivore effects were only detected on granite, where paradoxically, herbivore effects on vegetation are much less evident (Asner et al. 2009). The bulk density effects are of significance because they influence nutrient pools. Even when two sites contain similar nutrient concentrations on a mass basis, the site with higher bulk density has higher

© 2014 The Authors. Journal of Ecology © 2014 British Ecological Society, Journal of Ecology, 102, 1171–1182

1180 R. M. Holdo & M. C. Mack nutrient availability on a per-area basis for a given depth, and therefore a higher nutrient concentration within the rooting zone (although we must note that bulk density may also potentially affect plant nutrient uptake ability, and therefore it is unclear to what extent concentration differences capture availability differences). Previous studies have largely ignored site variation in bulk density (Belsky et al. 1993; Coetsee, February & Bond 2008; Treydte, Grant & Jeltsch 2009), yet we show that it varies as a function of some of the key drivers in this system. Our study, in conjunction with previous work in KNP (Wigley et al. 2013), underscores the fact that bulk density can vary substantially across space, with important implications for our understanding of soil fertility. SOIL TEMPERATURE AND MOISTURE DYNAMICS

Previous work conducted in Kenyan savannas (Belsky et al. 1989) found positive net effects of tree canopies on soil moisture. We found that the overall net effect of tree canopies on soil moisture was negligible over the course of the season. We found stronger patterns during and immediately following rainfall events. We found opposing effects of tree canopies on inputs on the two soil types: tree cover was associated with higher water inputs in the top 10 cm as a result of precipitation events on granite compared to open areas, but lower input rates on basalt. This suggests that tree canopies enhance water inputs on granite (for example via the process of stem flow, which may concentrate incoming precipitation near the tree trunk), but reduce them on basalt, most likely due to interception. To reinforce this conclusion, we computed the difference between the maximum and mean daily values of GWC for each sensor in our data set. We interpret this as the size of the initial input in soil moisture that occurs as a result of rainfall events. We then calculated the quantity DGWC by subtracting the former quantity for B sites from U sites for any given tree (only for days in which both sites had data). We treat DGWC as an approximate estimate of the magnitude of the ‘canopy’ effect on soil moisture input. We calculated mean DGWC values across all trees and regressed then against rainfall event sizes (including days with no rainfall). Our results were consistent with those of the state-space model: at the granite site, canopies are positively associated with inputs, and the size of the effect increased as a function of rainfall (Fig. S4 and Table S8). At the basalt site, the effect was weak, but statistically significant: canopies tended to slightly reduce water inputs in a nonlinear fashion (Fig. S4 and Table S8). As rainfall increased, U sites had progressively lower inputs, but the effect disappeared at higher rainfall. The latter results are consistent with interception effects. It is difficult at this point to fully explain the contrasting patterns at the two sites, given that they differ in multiple ways (not only geology but plant species composition and climate). The basalt site is significantly drier and hotter than the granite site, and our results are consistent with the observation that canopy interception and evaporation are expected to play a more important role on moisture inputs in arid than mesic sites (Rodriguez-Iturbe et al. 1999). The two sites also differ

in terms of tree species composition: our granite site was dominated by T. sericea and Schlerocarya birrea, while the basalt site is dominated by C. mopane and Combretum imberbe (Table S9). These two communities may differ in terms of leaf area index and architecture in ways that affect interception and stem-flow. We also considered the possibility of browsing differences contributing to the canopy effect, but we consider this unlikely. Our population of sites with U data consisted of trees inside the exclosure on basalt (where they are protected from browsing damage) and trees from both sides of the fence on granite. Browsing damage at the latter site is minimal (Asner et al. 2009), however, and therefore we consider it unlikely that browsing damage on the subsample of trees at this site explain the differences between the two sites. Despite the uncertainty about the exact mechanisms at play, we are fairly confident that the input differences are driven by canopy-related effects on water entering the system, and not by differences in water use, given the rapid time scales (sub-daily) over which they occur. Our results also suggest that herbivores exert greater overall effects on soil moisture dynamics than on plant-available nutrients, based on the fact that soil moisture effects were detected while labile nutrient effects were absent. The soil moisture model suggested that high herbivore biomass is associated with smaller rates of fast output following precipitation (and therefore a tendency for incoming moisture to remain in the top 10 cm of soil for longer than would be the case without herbivores), which, following the structure of our model, would lead us to suggest that differences in drainage primarily explain the association between herbivores and soil moisture availability. A potential explanation for this pattern is that it is driven by differences in soil bulk density, with denser soils having a lower saturated conductivity, and therefore lower infiltration rates. On granite at least, there is a clear pattern of increasing bulk density from canopy to intercanopy and across the exclosure fence, matching soil moisture patterns. We observe the soil moisture effect of the exclosure even after accounting for inter-site variation in bulk density, however, suggesting that other unknown factors may be at play. Our model found no effects of any of the covariates on slow output (which we mechanistically associate primarily with ET) at either site (Table 2). Given that our estimated coefficients for this process (e.g. b0) had very wide credible intervals (Tables S5 and S6), we believe that the model has low power when it comes to estimating ET effects on soil moisture dynamics. It is conceivable therefore that our herbivore effect may not be a drainage effect, but rather an ET effect that arises from differential grazing on transpiring vegetation (or differences in grass community composition) outside the exclosure. We believe that this is unlikely, however, because the herbivore effects are manifested primarily immediately following rainfall events, when drainage is the dominant mechanisms drawing down soil moisture. Overall, our study suggests that: canopy cover is a dominant biotic driver of soil C, N and P dynamics in these savanna systems; herbivores appear to affect soil moisture availability to a greater extent than they do soil nutrients; and

© 2014 The Authors. Journal of Ecology © 2014 British Ecological Society, Journal of Ecology, 102, 1171–1182

Herbivore and tree effects on savanna: soils 1181 canopy and herbivore effects on soil nutrient availability tend to be additive, given the lack of significant interaction effects in our analysis. We suggest that herbivores may shift limitation towards nutrients by slowing down soil moisture drainage from the upper soil layer following precipitation events. Finally, we propose that potentially important canopy effects on soil moisture turnover may be revealed by a closer examination of the various mechanisms that control soil moisture dynamics, even when time-averaged effects are small.

Acknowledgements We thank SANParks for allowing access to Kruger NP. We acknowledge Navashni Govender and Rina Grant from Scientific Services and the SANParks staff for assistance with field logistics. Garrett Arnold assisted with field and laboratory work, and Martin Lavoie, Jenny De Marco, Grace Crummer, Julia Reiskind, and Vanessa Trujillo provided additional help in the lab. Two anonymous referees provided helpful comments on a previous version of the manuscript. This research was funded by a grant from the Andrew W. Mellon Foundation. The authors declare that they have no conflict of interest.

Data accessibility Data available from the Dryad Digital Repository http://dx.doi.org/10.5061/ dryad.59k81 (Holdo & Mack 2014).

References Archibald, S., Bond, W.J., Stock, W.D. & Fairbanks, D.H.K. (2005) Shaping the landscape: fire–grazer interactions in an African savanna. Ecological Applications, 15, 96–109. Asner, G.P., Levick, S.R., Kennedy-Bowdoin, T., Knapp, D.E., Emerson, R., Jacobson, J., Colgan, M.S. & Martin, R.E. (2009) Large-scale impacts of herbivores on the structural diversity of African savannas. Proceedings of the National Academy of Sciences of the USA, 106, 4947–4952. Augustine, D.J., McNaughton, S.J. & Frank, D.A. (2003) Feedbacks between soil nutrients and large herbivores in a managed savanna ecosystem. Ecological Applications, 13, 1325–1337. Belsky, A.J. (1994) Influences of trees on savanna productivity – tests of shade, nutrients, and tree-grass competition. Ecology, 75, 922–932. Belsky, A.J. & Canham, C.D. (1994) Forest gaps and isolated savanna trees – an application of patch dynamics in two ecosystems. BioScience, 44, 77–84. Belsky, A.J., Amundson, R.G., Duxbury, J.M., Riha, S.J., Ali, A.R. & Mwonga, S.M. (1989) The effects of trees on their physical, chemical, and biological environments in a semi-arid savanna in Kenya. Journal of Applied Ecology, 26, 1005–1024. Belsky, A.J., Mwonga, S.M., Amundson, R.G., Duxbury, J.M. & Ali, A.R. (1993) Comparative effects of isolated trees on their undercanopy environments in high-rainfall and low-rainfall savannas. Journal of Applied Ecology, 30, 143–155. Biondini, M.E. & Manske, L. (1996) Grazing frequency and ecosystem processes in a northern mixed prairie, USA. Ecological Applications, 6, 239–256. Bobbink, R., Hicks, K., Galloway, J., Spranger, T., Alkemade, R., Ashmore, M. et al. (2010) Global assessment of nitrogen deposition effects on terrestrial plant diversity: a synthesis. Ecological Applications, 20, 30–59. Bond, W.J. (2005) Large parts of the world are brown or black: a different view on the ‘Green World’ hypothesis. Journal of Vegetation Science, 16, 261–266. Bond, W.J., Woodward, F.I. & Midgley, G.F. (2005) The global distribution of ecosystems in a world without fire. New Phytologist, 165, 525–537. Brudvig, L.A. & Asbjornsen, H. (2009) The removal of woody encroachment restores biophysical gradients in Midwestern oak savannas. Journal of Applied Ecology, 46, 231–240. Chapin, F.S. & McNaughton, S.J. (1989) Lack of compensatory growth under phosphorus deficiency in grazing-adapted grasses from the Serengeti Plains. Oecologia, 79, 551–557. Clark, J.S. & Bjornstad, O.N. (2004) Population time series: process variability, observation errors, missing values, lags, and hidden states. Ecology, 85, 3140–3150.

Coetsee, C., Bond, W.J. & February, E.C. (2010) Frequent fire affects soil nitrogen and carbon in an African savanna by changing woody cover. Oecologia, 162, 1027–1034. Coetsee, C., February, E.C. & Bond, W.J. (2008) Nitrogen availability is not affected by frequent fire in a South African savanna. Journal of Tropical Ecology, 24, 647–654. Coughenour, M.B. (1984) A mechanistic simulation analysis of water use, leaf angles, and grazing in East African graminoids. Ecological Modelling, 26, 203–230. Craine, J.M., Morrow, C. & Stock, W.D. (2008) Nutrient concentration ratios and co-limitation in South African grasslands. New Phytologist, 179, 829–836. Cumming, D.H.M. & Cumming, G.S. (2003) Ungulate community structure and ecological processes: body size, hoof area and trampling in African savannas. Oecologia, 134, 560–568. Dean, W.R.J., Milton, S.J. & Jeltsch, F. (1999) Large trees, fertile islands, and birds in arid savanna. Journal of Arid Environments, 41, 61–78. R Development Core Team (2011) R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. February, E.C., Higgins, S.I., Bond, W.J. & Swemmer, L. (2013) Influence of competition and rainfall manipulation on the growth responses of savanna trees and grasses. Ecology, 94, 1155–1164. Frank, D.A. & Groffman, P.M. (1998) Ungulate vs. landscape control of soil C and N processes in grasslands of Yellowstone National Park. Ecology, 79, 2229–2241. Frank, D.A., Inouye, R.S., Huntly, N., Minshall, G.W. & Anderson, J.E. (1994) The biogeochemistry of a north-temperate grassland with native ungulates – nitrogen dynamics in Yellowstone National Park. Biogeochemistry, 26, 163–188. Guswa, A.J., Celia, M.A. & Rodriguez-Iturbe, I. (2002) Models of soil moisture dynamics in ecohydrology: a comparative study. Water Resources Research, 38, 1–15. Hoffmann, W.A. (1999) Fire and population dynamics of woody plants in a neotropical savanna: matrix model projections. Ecology, 80, 1354–1369. Holdo, R.M. (2007) Elephants, fire, and frost can determine community structure and composition in Kalahari woodlands. Ecological Applications, 17, 558–568. Holdo, R.M. & Mack, M.C. (2014) Data from: functional attributes of savannah soils: contrasting effects of tree canopies and herbivores on bulk density, nutrients and moisture dynamics. Dryad Digital Repository, http://dx.doi.org/ 10.5061/dryad.59k81. Holdo, R.M., Mack, M.C. & Arnold, S.G. (2012) Tree canopies explain fire effects on soil nitrogen, phosphorus and carbon in a savanna ecosystem. Journal of Vegetation Science, 23, 352–360. Holdo, R.M., Sinclair, A.R.E., Metzger, K.L., Bolker, B.M., Dobson, A.P., Ritchie, M.E. & Holt, R.D. (2009) A disease-mediated trophic cascade in the Serengeti and its implications for ecosystem C. PLOS Biology, 7, e1000210. Johnson, L.C. & Matchett, J.R. (2001) Fire and grazing regulate belowground processes in tallgrass prairie. Ecology, 82, 3377–3389. Knoop, W.T. & Walker, B.H. (1985) Interactions of woody and herbaceous vegetation in a southern African savanna. Journal of Ecology, 73, 235–253. Kuo, S. (1996) Phosphorus. Methods for Soil Analysis – Part 3 – Chemical Methods (eds D.L. Sparks, A.L. Page, P.A. Helmke, R.H. Loeppert & P.N. Soltanpour), pp. 869–920. Soil Science Society of America, Madison, WI. Laio, F., Porporato, A., Ridolfi, L. & Rodriguez-Iturbe, I. (2001) Plants in water-controlled ecosystems: active role in hydrologic processes and response to water stress – II. Probabilistic soil moisture dynamics. Advances in Water Resources, 24, 707–723. Lavoie, M., Mack, M.C. & Schuur, E.A.G. (2011) Effects of elevated nitrogen and temperature on carbon and nitrogen dynamics in Alaskan arctic and boreal soils. Journal of Geophysical Research, 116, G03013. Legendre, P. & Legendre, L. (1998) Numerical Ecology. Elsevier, Amsterdam. Levick, S.R., Asner, G.P., Kennedy-Bowdoin, T. & Knapp, D.E. (2009) The relative influence of fire and herbivory on savanna three-dimensional vegetation structure. Biological Conservation, 142, 1693–1700. Levick, S.R., Asner, G.P., Kennedy-Bowdoin, T. & Knapp, D.E. (2010) The spatial extent of termite influences on herbivore browsing in an African savanna. Biological Conservation, 143, 2462–2467. Ludwig, F., Dawson, T.E., Kroon, H., Berendse, F. & Prins, H.H.T. (2003) Hydraulic lift in Acacia tortilis trees on an East African savanna. Oecologia, 134, 293–300. Ludwig, F., de Kroon, H., Berendse, F. & Prins, H.H.T. (2004) The influence of savanna trees on nutrient, water and light availability and the understorey vegetation. Plant Ecology, 170, 93–105.

© 2014 The Authors. Journal of Ecology © 2014 British Ecological Society, Journal of Ecology, 102, 1171–1182

1182 R. M. Holdo & M. C. Mack McCarthy, M.A. (2007) Bayesian Methods for Ecology. Cambridge University Press, Cambridge. McNaughton, S.J. (1988) Mineral nutrition and spatial concentrations of African ungulates. Nature, 334, 343–345. McNaughton, S.J., Banyikwa, F.F. & McNaughton, M.M. (1997) Promotion of the cycling of diet-enhancing nutrients by African grazers. Science, 278, 1798–1800. Milchunas, D.G. & Lauenroth, W.K. (1993) Quantitative effects of grazing on vegetation and soils over a global range of environments. Ecological Monographs, 63, 327–366. Naiman, R.J., Braack, L., Grant, R., Kemp, A.C., du Toit, J.T. & Venter, F.J. (2003) Interactions between species and ecosystem characteristics. The Kruger Experience: Ecology and Management of Savanna Heterogeneity (eds J.T. Du Toit, K.H. Rogers & H.C. Biggs), pp. 221–241. Island Press, Washington, DC. Parton, W.J., Schimel, D.S., Cole, C.V. & Ojima, D.S. (1987) Analysis of factors controlling soil organic-matter levels in Great-Plains grasslands. Soil Science Society of America Journal, 51, 1173–1179. Penning de Vries, F., Krul, J. & van Keulen, H. (1980) Productivity of Sahelian rangelands in relation to the availability of nitrogen and phosphorus from the soil. Nitrogen Cycling in West African Ecosystems (ed. T. Rosswall), pp. 95–113. Royal Swedish Academy of Science, Stockholm. Pinheiro, J.C. & Bates, M. (2000) Mixed-effects Models in S and S-PLUS. Springer-Verlag, New York. Riginos, C. (2009) Grass competition suppresses savanna tree growth across multiple demographic stages. Ecology, 90, 335–340. Rodriguez-Iturbe, I., Porporato, A., Ridolfi, L., Isham, V. & Cox, D.R. (1999) Probabilistic modelling of water balance at a point: the role of climate, soil and vegetation. Proceedings: Mathematical Physical and Engineering Sciences, 455, 3789–3805. Rodr!ıguez-Iturbe, I., D’Odorico, P., Porporato, A. & Ridolfi, L. (1999) Tree-grass coexistence in savannas: the role of spatial dynamics and climate fluctuations. Geophysical Research Letters, 26, 247–250. Ruess, R.W. & Seagle, S.W. (1994) Landscape patterns in soil microbial processes in the Serengeti National Park, Tanzania. Ecology, 75, 892–904. Sankaran, M., Ratnam, J. & Hanan, N. (2008) Woody cover in African savannas: the role of resources, fire and herbivory. Global Ecology and Biogeography, 17, 236–245. Scholes, R.J. & Archer, S.R. (1997) Tree–grass interactions in savannas. Annual Review of Ecology and Systematics, 28, 517–544. Scholes, R.J. & Walker, B.H. (1993) An African Savanna: Synthesis of the Nylsvlei Study. Cambridge University Press, Cambridge. Schulze, E.D., Caldwell, M.M., Canadell, J., Mooney, H.A., Jackson, R.B., Parson, D., Scholes, R., Sala, O.E. & Trimborn, P. (1998) Downward flux of water through roots (ie inverse hydraulic lift) in dry Kalahari sands. Oecologia, 115, 460–462. Staver, A.C., Archibald, S. & Levin, S. (2011) Tree cover in sub-Saharan Africa: rainfall and fire constrain forest and savanna as alternative stable states. Ecology, 92, 1063–1072. Staver, A.C., Bond, W.J., Stock, W.D., van Rensburg, S.J. & Waldram, M.S. (2009) Browsing and fire interact to suppress tree density in an African savanna. Ecological Applications, 19, 1909–1919. du Toit, J.T. (2003) Large herbivores and savanna heterogeneity. The Kruger Experience: Ecology and Management of Savanna Heterogeneity (eds J.T. Du Toit, K.H. Rogers & H.C. Biggs), pp. 292–309. Island Press, Washington, DC. Treydte, A.C., Grant, C.C. & Jeltsch, F. (2009) Tree size and herbivory determine below-canopy grass quality and species composition in savannahs. Biodiversity and Conservation, 18, 3989–4002. Treydte, A.C., van der Beek, J.G.M., Perdok, A.A. & van Wieren, S.E. (2010) Grazing ungulates select for grasses growing beneath trees in African savannas. Mammalian Biology – Zeitschrift fur Saugetierkunde, 76, 345–350. Venter, F.J., Scholes, R.J. & Eckhardt, H.C. (2003) The abiotic template and its associated vegetation pattern. The Kruger Experience: Ecology and Management of Savanna Heterogeneity (eds J. Du Toit, K.H. Rogers & H. Biggs), pp. 83–129. Island Press, Washington, DC. Vetaas, O.R. (1992) Micro-site effects of trees and shrubs in dry savannas. Journal of Vegetation Science, 3, 337–344. Vitousek, P.M. & Howarth, R.W. (1991) Nitrogen limitation on land and sea: how can it occur? Biogeochemistry, 13, 87–115. Whyte, I.J., van Aarde, R.J. & Pimm, S.L. (2003) Kruger’s elephant population: it’s size and consequences for ecosystem heterogeneity. The

Kruger Experience (eds J. Du Toit, K.H. Rogers & H. Biggs), pp. 332–348. Island Press, Washington, DC. Wigley, B.J., Coetsee, C., Hartshorn, A.S. & Bond, W.J. (2013) What do ecologists miss by not digging deep enough? Insights and methodological guidelines for assessing soil fertility status in ecological studies. Acta Oecologica, 51, 17–27. van Wilgen, B.W., Trollope, W.S.W., Biggs, H.C., Potgieter, A.L.F. & Brockett, B.H. (2003) Fire as a driver of ecosystem variability. The Kruger Experience: Ecology and Management of Savanna Heterogeneity (eds J.T. Du Toit, K.H. Rogers & H.C. Biggs), pp. 149–170. Island Press, Washington, DC. Received 23 December 2013; accepted 24 June 2014 Handling Editor: Susan Schwinning

Supporting Information Additional Supporting Information may be found in the online version of this article: Figure S1. Analysis of spatial autocorrelation among sampling points in the herbivore exclosures. Figure S2. Sample segment of soil moisture time series. Figure S3. Soil bulk density as a function of total C at the granite site. Figure S4. Canopy effects on short-term soil moisture inputs on granite and basalt. Table S1. ANOVA tables corresponding to the summary results for granite in Table 1. Table S2. ANOVA tables corresponding to the summary results for basalt in Table 1. Table S3. Soil temperature results for granite. Table S4. Soil temperature results for basalt. Table S5. Full parameter estimates for the soil moisture model based on site-specific bulk density values. Table S6. Full parameter estimates for the soil moisture model based on mean bulk density values. Table S7. True and inferred soil moisture dynamics parameter values for a simulated data set. Table S8. ANOVA tables for canopy effects on short-term soil moisture inputs on granite and basalt. Table S9. Species and crown diameters of trees used in the soil moisture dynamics and temperature studies. Appendix S1. Description and model code for soil moisture statespace model.

© 2014 The Authors. Journal of Ecology © 2014 British Ecological Society, Journal of Ecology, 102, 1171–1182