1. Termites and large herbivores represent important functional groups in savanna ecosystems. Termites affect vegetation far beyond their mounds. In addition, large herbivores feed selectively on termite mound vegetation or in the vicinity of mounds. Previous studies of savanna vegetation communities have focused on termites and large herbivores separately, although interaction effects may be predicted.
2. We studied the effects of large herbivores and large vegetated Macrotermes mounds on the herbaceous vegetation in Lake Mburo National Park in Uganda. We recorded herbaceous vegetation change over 3 years on savanna areas (with and without large herbivores) and on corresponding termite mounds (with and without large herbivores) in a randomized block design.
3. Termite mounds and savannas had significantly different plant communities, but large herbivore grazing exclusions did not result in significant shifts in plant communities during this study period. A canonical correspondence analysis separated species mainly along an axis from termitaria to savanna. Only a few species responded to grazing exclusion. Some erect species, such as Hyparrhenia filipendula and Themeda triandra, increased in cover, and creeping species, such as Cynodon dactylon, decreased, following the exclusion of grazers. Forbs dominated mound areas, while graminoids dominated the savanna areas. Fencing increased the cover of graminoids over time and led to gradual increase in the relative cover of graminoids compared with forbs.
4. Mound soil was higher in pH, calcium and magnesium and lower in sodium compared with adjacent savanna areas. Nitrogen and carbon soil content did not differ between the two habitats. Soil phosphorus increased following grazing exclusion.
5.Synthesis. This study shows that termites may exert a far more important effect on the herbaceous community than large herbivores in savanna areas, even if the biomass of large herbivores is relatively high. Thus, future studies on savanna vegetation ecology should focus increasingly on important insect groups in addition to the more conspicuous large mammal guild.
Both termites and large herbivores are known to redistribute considerable amounts of nutrients and consequently generate nutrient hotspots in savannas (Fox-Dobbs et al. 2010; Stock, Bond & van de Vijver 2010). In east Africa, Fox-Dobbs et al. (2010) found that termite mound soil had higher levels of nitrogen and that termites had a greater effect on nitrogen fixation than large mammals. Grazed and ungrazed plots had no consistent differences in nitrogen (N), phosphorus (P) or N/P ratio in the Serengeti, but the relative plant species richness decreased at sites where the exclusion of herbivores increased available P (Anderson, Ritchie & McNaughton 2007). In Kenyan rangelands, large herbivores caused a net input of nitrogen (N) at nutrient-rich sites and a net loss from nutrient-poor sites (Augustine, McNaughton & Frank 2003). Imperata cylindrica-dominated grassland in lowland Nepal intensively grazed by axis deer (Axix axis) had reduced soil P content compared with adjacent ungrazed grassland (Moe & Wegge 2008).
In the flat valley areas of Lake Mburo National Park, Uganda, large termite mounds (termitaria) built by Macrotermes species support diverse plant assemblages (Bloesch 2008). Although occupying only about 5% of the landscape in the flat valley areas, termitaria explain up to 89% of the variation in thicket clumps (Moe, Mobæk & Narmo 2009). In a recent study from South Africa, Levick et al. (2010) found that termites influenced vegetation cover as much as 20% at the landscape scale and Pringle et al. (2010) concluded that termites were of major importance for savanna ecosystem function. Vegetated termitaria are preferred grazing, browsing and resting sites for large- and medium-size ungulates (Mobæk, Narmo & Moe 2005). Therefore, both termites and mammalian herbivores may alter the relative dominance of grasses and forbs, the two most common functional plant groups in savanna grasslands. An earlier study from Lake Mburo National Park has shown that forbs dominate the vegetated Macrotermes mounds, and grasses tend to dominate the adjacent savanna areas (Moe, Mobæk & Narmo 2009). Grazing by large herbivores may increase grass species dominance over forbs (Rooney 2009), although in Uruguay, Rodríguez et al. (2003) concluded that the forb dynamics were regulated by other processes than grazing. Despite potential interaction effects (Fox-Dobbs et al. 2010) on savanna vegetation, previous studies have focused on large mammals and termites separately. This study focused on how large herbivores and termites may act in concert to affect savanna herbaceous plant communities. To our knowledge, such studies have not been conducted previously.
In this study, our goal was to assess how herbaceous species composition on termitaria and adjacent savanna change over time when large herbivores are excluded. We predicted that both large herbivores and termitaria affect herbaceous species composition. We also postulated that excluding large herbivores would have less effect on herbaceous species composition on savanna areas than excluding large herbivores on termitaria because (i) theoretical and empirical evidence suggests that the effect of grazing on plant community composition increases with habitat productivity and termitaria are relatively resource rich (Milchunas, Sala & Lauenroth 1988; Chase et al. 2000; Osem, Perevolotsky & Kigel 2004) and (ii) large herbivores preferentially graze on termite mound vegetation (Mobæk, Narmo & Moe 2005). We also predicted that termitaria, rather than grazing, determine the changes in the relative cover of graminoids and forbs because functional groups are predicted to exhibit life-form-specific limitations to different abiotic and biotic constraints within plant communities (Shaver & Chapin 1980; Chapin & Shaver 1985; Osem, Perevolotsky & Kigel 2004; Bezemer et al. 2006), which are often attributed to differences in growth form, developmental constraints and physiology (Chapin 1980; Aerts 1999). Finally, we predicted that soil nutrient content may be higher on termitaria than on adjacent savanna soil (Bloesch 2008; Fox-Dobbs et al. 2010) and that herbivore exclusion would alter soil nutrient content (Augustine, McNaughton & Frank 2003).
Materials and methods
This study was conducted in Lake Mburo National Park, located in south-western Uganda. The park covers approximately 260 km2 and lies close to the equator (00°30′, 00°45′ S, 45°00′, 31°05′ E) at an elevation of about 1200 m above sea level.The study sites were located in the base of flat valleys found in the eastern and western part of the park. The flat valley areas are characterized by homogenous landscapes punctuated with vegetated termitaria thickets constructed mainly by large mound-building Macrotermes termite species (Bakuneeta 1989). The soils of the park are mainly ferrasols, histosols, vertisols and leptosols (Bloesch 2002). The average annual rainfall is c. 800 mm per year with two peaks occurring in February and October. The driest months are July and August. The average annual temperature is c. 27.5 °C (Bloesch 2002). The vegetation of the park is composed of mixed woodlands interspersed with thicket clumps. The most common species in the woodlands are Acacia hockii, A. gerrardii, Rhus natalensis and Grewia spp., while Sporobolus pyramidalis and Themeda triandra comprise the important grassland vegetation (Moe, Mobæk & Narmo 2009).
The biomass density of wild unglates within the park is estimated to be 86.64 kg ha−1 (Rannestad et al. 2006). Common ungulates in the park are impala (Aepyceros melampus), African buffalo (Syncerus caffer), Burchell’s zebra (Equus burchelli), bushbuck (Tragelaphus scriptus) and warthogs (Phacochoerus aethiopicus) (Rannestad et al. 2006).
Nine sites were randomly selected within a 260-km2 area between June and July 2005. At each site, two adjacent termite mounds and two adjacent savanna areas (i.e. four plots) were selected. Subsequently, one of the mound plots and one of the savanna plots were randomly selected and fenced (Fig. 1). Sizes of the plots varied from 90 to 260 m2 depending on the size of the mounds at each site. The four plots were always the same size at each site. Only termite mounds that were active at the onset of the experiment were included. No active or inactive mounds were present in the savanna plots. Macrotermes mounds are found throughout the park with densities of 10.1 per ha (Moe, Mobæk & Narmo 2009). The radii of experimental mounds varied from 2.38 to 5.13 m with a mean of 3.23 m (±0.71 SD, n =18). The average height of the termite mounds was 1.83 m (±0.46 SD). The tree canopy cover on termitaria thickets varied from 39.02 to 162.78 m2 with a mean of 80.48 (±34.73 SD). The intermound distances varied from 22 to 53 m with a mean of 35 m (±11.33 SD, n =18). Plots within a site were an average distance of 35.22 m (±11.33 SD) apart, far enough (>20 m) to minimize the risk of savanna plots being close to other mounds (Loveridge & Moe 2004; Levick et al. 2010). Fences were 2 m high and made of galvanized chain links with 5-cm mesh, supported by 2-m high steel angle-bars, and firmly fixed to the ground with concrete to prevent damage by large herbivores.
We used the point intercept method (PIM) to sample herbaceous vegetation (Goodall 1953). Four permanent 40 × 50 cm quadrats were established in each plot. The position of each quadrat was permanently marked with small aluminium pins to ensure accurate relocation each time. The quadrats were randomly located in savanna plots (at least 1 m from the fenced edge). To ensure representative sampling on the elevated termitaria, quadrats were located at the four locations defined as top, middle, base (where the mound levelled off) and edge (1 m from the base). The point intercept table was placed above each quadrat and a 1.8-mm-diameter pin was inserted vertically through 20 points spread uniformly on the grid (Levy & Madden 1933; Goodall 1953), giving a total of 80 points per plot (20 points × four quadrats). All contacts with the pins were recorded (i.e. also when the same individual was in contact with the pin more than once) (Frank & McNaughton 1990). The cover of each species was estimated by recording the number of contacts each species made with the pin (Goodall 1953). Foliar contacts were recorded for graminoids and forbs. Baseline sampling was made before fencing in early July 2005, and then, subsequent assessments were made in October and November 2005, March and April 2006, July and August 2006, October and November 2006, March and April 2007, July and August 2007, and August 2008 (i.e. eight sampling sessions). Sampling sessions were grouped into dry season (July and August) and wet season (October, November, March and April). All plants were identified to species level. If a species could not be identified in the field, it was taken to the Makerere University Herbarium for identification. Plant identification followed the Flora of Tropical East Africa (Beentje & Cheek 2003).
Two and a half years after the onset of the experiment (February 2008), the nutrient properties of the soil were assessed by collecting soil cores from the top 15 cm of soil. Topsoil represents nutrient availability for herbaceous vegetation in arid and semi-arid African savannas, as their roots are mainly located in this layer (Mordelet, Menaut & Mariotti 1997; February & Higgins 2010). Four samples per plot were taken with a hoe and combined into one sample. Samples were air-dried in the field and transported to the laboratory for further processing. The analyses were conducted at the Eurofins laboratory, Moss, Norway. The soil samples were crushed and passed through a 2-mm sieve. The <2-mm soil fraction was analysed according to the procedure of (Egner, Riehm & Domingo 1960), in which exchangeable calcium (Ca), magnesium (Mg), sodium (Na) and potassium (K) were extracted with 100 mL of 0.1 M ammonium lactate + 0.4 M acetic acid (pH 3.75). Na and K were determined by a flame photometer; Ca and Mg by atomic absorption spectrophotometry; and phosphorus by colorimetric technique. Organic carbon (C) was determined by the Walkley–Black method (Walkley & Black 1934), and total nitrogen (N) by the Kjeldahl method. In addition, we analysed pH using a pH meter, because pH affects the availability and uptake of other nutrients for plants (Chapin 1980).
To explore how termites and large herbivore grazing influence the herbaceous species composition, we used ordination methods. We chose canonical correspondence analysis (CCA) because the initial exploration of the data using detrended correspondence analysis (DCA) indicated that the length of gradients in the first axis was long (4.7 SD). Long gradient length in DCA suggests that species were responding nonlinearly to the gradients of explanatory variables, and therefore, unimodal response models were appropriate (Lepš & Šmilauer 2003). Data from the four permanent 40 × 50 cm quadrats in each plot were combined in the statistical analyses to avoid spatial pseudoreplication. We removed infrequent species with fewer than three occurrences from the data set before analysis, a common method in the analysis of gradients in communities (McCune, Grace & Urban 2002). Forward selection was used to select habitat type, pH, Ca, Mg, K, N and total C that significantly explained the variance in the species composition data. Explanatory variables with P >0.05, such as K, Na and P, were excluded from further CCA analysis.
Variance partitioning (Borcard, Legendre & Drapeau 1992) was used to assess how the different variables contributed to the explanation of species composition. The CCA runs were constrained by (i) habitat type, (ii) grazing, (iii) soil variables and (iv) season. Each variable was run with the remaining three variables as covariables.
Changes in the species composition in the treatments over time were analysed using the principal response curve (PRC) method. The PRC is a partial form of the constrained linear ordination method of redundancy analysis (Van den Brink & Ter Braak 1999). We chose the PRC method because it overcomes the difficulties posed in interpreting temporal trends in ordination plots and permits the detection of treatment effects at the species level. The PRC plots treatment effects against time, expressing the effects of treatment as deviations from the control treatment (van den Brink et al. 2009). We used the open savanna treatment as a control. The PRC was applied with time of sampling (Time) and treatment – savanna open, savanna fenced, termitaria open and termitaria fenced – as categorical dummy variables to test the null hypothesis that large herbivores and termitaria have no effect on herbaceous species composition. Associated with each PRC is the species weight (bk), which shows the change in the cover of each species in the treatments relative to the control. Effect sizes are calculated for each treatment using the species weights and treatment scores (Ter Braak & Smilauer 2003). The null hypothesis assumes that the PRC diagram does not capture the treatment variance (i.e. Cdt × bk = 0 for t, d and k). Species with high species weights (bk) follow the overall community response and are highly affected by the treatment.
The level of significance of the PRC axis and the time of sampling, for which a significant difference was apparent between treatments, was tested under a reduced model while keeping the same plots sampled at different times together (restricted for split plots) and permuting the plot (whole plots) in Canoco terms (see Fig. 1) (Ter Braak & Smilauer 2003). All DCA, CCA and PRC were performed by Canoco for Windows version 4.5, and ordination diagrams were drawn in Canoco Draw (Ter Braak & Smilauer 2003). Significance of variables was tested using a Monte Carlo permutation test with 499 permutations, which is adequate for a test at the 5% significance level. In all analyses, species cover data were ln(2x + 1)-transformed following Van den Brink et al. (2000).
To test our hypothesis that excluding herbivores will have a weaker effect on species composition in savanna compared with termitaria (how similar treatment effects were during the whole study period), we used the function ‘adonis’ in the vegan library in R (R Development Core Team 2011) (Oksanen et al. 2010), based on Bray–Curtis dissimilarity values of species cover. To compare the differences in species composition among treatments, we used the function ‘betadisper’ in the vegan library in R (Oksanen et al. 2010) following Anderson’s (2001) approach. The functions adonis and betadisper are multivariate analogues of anova and Levene’s test for comparing group means and variances, respectively. We tested pairwise differences between treatments using parametric Tukey’s HSD test and assessed the 95% confidence intervals around treatment centroids to show differences in community composition. We tested whether there were differences in community composition as a function of sampling date using a multiresponse permutation procedure (MRPP), a nonparametric randomization-based alternative to multivariate anova.
To explore the effects of termitaria and large herbivores on the cover of common individual herbaceous species over time, we used generalized linear mixed models (lmer in the library lme4 in R version 2.12.0, R Development Core Team (2011). We used Poisson errors and log link functions because the response variable ‘cover’ represented a number of hits (Crawley 2007).
To address our hypothesis that the shift from graminoid cover to forbs was induced by termitaria rather than grazing, we used linear mixed models (lme in the library nlme in R version 2.12.0). We used habitat type, grazing, season, time and time2 as fixed factors. Plots were used as random variables to account for the dependency in data point owing to repeated surveys (Crawley 2007). Because response variables (cover) were counts (number of pin hits), we ln-transformed the data to improve the homogeneity of variance (Crawley 2007). We selected the optimal random component and appropriate error structure of the model with a restricted maximum-likelihood (REML) algorithm (Zuur et al. 2009). When comparing models with different fixed terms, we used maximum likelihood because models fitted with REML cannot be compared with likelihood ratios (Pinheiro & Bates 2000). We chose the best model based on the Akaike’s Information Criterion (AIC) using the mixed-effect modelling.
Altogether, we identified 117 herbaceous species, of which 20 most common species by treatment are listed in the appendix (see Appendix S1 in Supporting Information). Habitat type (termitaria vs. savanna) largely determined the differences in species composition (Appendix S1, Fig. 2). Whether the area was grazed or not did not explain differences in overall species composition (Table 1). Termitaria were dominated by forbs; savanna areas were dominated by graminoids (Table 1, Fig. 2). Fencing increased graminoid cover over time and led to a gradual increase in the relative cover of graminoids but did not change the cover of forbs (Table 2). Some grass species, such as Sporobolus africanus, Brachiaria decumbens, Hyparrhenia filipendula and T. triandra, were common in both termitaria and savanna. Although forbs were largely exclusively associated with termitaria, Commelina africana, Dyschoriste nagchana, Indigofera spicata and Glycine wightii were found in both habitats.
Table 1. Differences in species composition detected using permutation tests for homogeneity of multivariate dispersions on Bray–Curtis distances for herbaceous community in four grazing treatments (savanna open, savanna fenced, termitaria open and termitaria fenced) (Tukey’s multiple comparisons of means with 95% family-wise confidence level)
Source of variation
SS, sum of square; MS, mean square.
Savanna open vs. savanna fenced
Termitaria open vs. termitaria fenced
Termitaria open vs. savanna open
Termitaria open vs. savanna fenced
Termitaria fenced vs. savanna fenced
Termitaria fenced vs. savanna open
Table 2. Parameter estimates for the minimum adequate model (selected with AIC) of the effect of habitat, grazing, season and time on the cover of graminoids and forbs and the ratio of graminoids to forbs. The model is a mixed-effect model where the cover is explained by habitat (savanna and termitaria), grazing (grazed and ungrazed), season (dry and wet) and time (continuous) and their interactions. The reference categories are savanna habitat, dry season and grazed areas. Response variables were log-transformed prior to analysis
AIC, Akaike’s information criterion.
Termitaria (vs. savanna)
Ungrazed (vs. grazed)
Wet (vs. dry)
Termitaria × wet
Ungrazed × time
Termitaria (vs. savanna)
Wet (vs. dry)
Termitaria × time
Wet × time
Termitaria (vs. savanna)
Ungrazed (vs. grazed)
Termitaria × time
Ungrazed × time
Wet × time
Although most abundant species were closely associated with the termitaria–savanna gradient, some species, such as Monothecium glandulosum, Setaria kagerensis and Asparagus racemosus, had their optimum clearly associated with Mg, and Teramnus labialis was associated with fenced savanna plots (Fig. 2). The forbs Pavonia patens and Cyanotis foecunda were mostly associated with unfenced termitaria and G. wightii with unfenced savanna areas (Fig. 2).
Savanna and termitaria were consistently different in species composition, irrespective of fencing treatment (Table 1). Although the species composition did not shift with time when areas were fenced, the tendency for shifts in species composition was greater in savanna (P =0.075) than in termitaria communities (P =0.971) (Table 1, Fig. 3).
The total variation in species composition explained by the first canonical axis of the PRC ordination was 72%. Of this total variation, 16.4% was attributed to treatment, while time of sampling explained only 7% (Fig. 3). Although overall grazing did not contribute much to differences in community species composition, the cover of some individual species, such as H. filipendula (P <0.001), T. triandra (P =0.028), Panicum maximum (P =0.003), Chloris gayana (P <0.001) and B. decumbens (P <0.001), increased significantly over time following the exclusion of grazers. The cover of the most common species S. africanus (P <0.001) decreased following large herbivore exclusion on both enclosed termitaria and savanna areas. Other species that decreased following exclusion of grazers were Cynodon dactylon (P <0.001), Panicum coloratum (P <0.001), C. africana (P =0.010) and I. spicata (P <0.001).
Termitaria were associated with significantly (P <0.001) higher pH (6.83 ± 0.52 SD) compared with adjacent savanna areas (5.96 ± 0.32 SD). Similarly, the levels of Mg and Ca were significantly (P <0.003 and P <0.0001, respectively) higher on termitaria. No differences in total C (P =0.733) or total N (P =0.07) were detected between termitaria and adjacent savanna plots. Grazing did not have a significant effect on pH (P =0.619) or the levels of Ca (330.25 mg 100 g−1± 120.01 SD, P =0.717) or Mg (72.89 ± 22.10 mg 100 g−1, P =0.492). The P levels were not different between savanna and termitaria habitats (6.44 ± 5.73, P =0.159); however, absence of grazing significantly (P <0.054) increased the levels of P in ungrazed sites compared with grazed sites.
Significant environmental variables altogether explained 89% of the total variance in species composition (Table 3). However, when environmental variables were grouped into habitat, grazing, season and soil, they differed considerably in their ability to explain species composition. Soil variables explained 44% of the variance, and habitat type explained only 15% (Table 3). In contrast, the contribution of grazing by large herbivores was not significant in explaining herbaceous species composition (P =0.23, Table 3).
Table 3. Variance partitioning by canonical correspondence analysis (CCA) showing percentage of unique contribution and maximum potential variance explained by habitat type, grazing, season and soil in the herbaceous species composition data. Unique contribution is the variance accounted for when the effect of all other variables is removed in partial CCA
Unique contribution (%)
Max. potential % variance explained
P values refer to the significance of unique contribution (P = probability estimated by Monte Carlo permutation tests).
pH, Ca, Mg, Kj N, total C
Habitat, grazing, season, soils
In previous studies, both termites and large herbivores have been found to affect savanna vegetation (Holdo & McDowell 2004; Moe, Mobæk & Narmo 2009). In our study, area termites, rather than herbivores, determined the herbaceous vegetation community. Species composition differed considerably between termitaria and adjacent savanna areas (Table 1, Fig. 2). Excluding large herbivores did not affect the overall species composition of savanna or termitaria habitats (Table 1, Fig. 3). Thus, our prediction that both large herbivores and termitaria independently affect herbaceous species composition was not supported. The finding that large termitaria exert a pronounced effect on plant species composition in arid and semi-arid African savanna grasslands is not surprising (Sileshi et al. 2010). However, the fact that the effect of herbivores on herbaceous species composition was consistent over time, despite large seasonal fluctuations and significant differences in species composition between termitaria and savanna, challenges our current knowledge of savanna ecology.
Lake Mburo National Park supports a wild ungulate biomass density comparable to many other African savanna areas (Coe, Cumming & Phillipson 1976; Mizutani 1999; Mandujano & Naranjo 2010). Our study monitored the whole herbaceous community in an experimental setting over 3 years, with a multivariate analysis approach sensitive to detecting community differences (Anderson 2001; Lepš & Šmilauer 2003). We found temporal changes that resulted in relatively large differences in species composition between termite mounds and savanna (Fig. 3). The plant composition differences peaked 16 months after the onset of the study. We believe that these changes can be attributed to variation in rainfall which coincided with these sampling sessions. Months 12 and 16 were in the El Niño period and month 24 corresponds to the dry period (La Niña) experienced in the region during the course of the study (Becker et al. 2010). Thus, termite mound vegetation and savanna vegetation may respond differently to variations in rainfall. However, despite such temporal differences between mound and savanna vegetation, there were no treatment differences caused by herbivores, supporting our hypothesis that termites are the main determinants of variations in savanna herbaceous communities in these areas.
Although we did not detect major effects of large herbivores on species composition, certain species did change in cover following the exclusion of large mammals. B. decumbens, a palatable graminoid, increased in the fenced areas. The graminoids H. filipendula and T. triandra increased in cover when fenced, although they are known to be tolerant to grazing by effectively using nutrients to recover quickly from defoliation (McNaughton 1985). As found in previous studies (e.g. McNaughton 1979; Díaz et al. 2007), there was a tendency for relatively tall, erect graminoids, such as H. filipendula and T. triandra, to increase in cover, while the cover of short, creeping graminoid species, such as C. dactylon, decreased. However, the cover of B. decumbens, a short, creeping graminoid species, did increase substantially after grazing exclusion.
We also predicted greater changes in herbaceous species composition on termitaria habitat than on adjacent savanna areas following herbivore exclusion, because plant composition shifts are expected to be higher at more productive sites (Milchunas, Sala & Lauenroth 1988; Chase et al. 2000; Osem, Perevolotsky & Kigel 2004), and large mammals strongly prefer to feed on more productive sites (Holdo & McDowell 2004; Mobæk, Narmo & Moe 2005). Our data did not support this predication. Instead, there was a tendency (P =0.075, Table 1) for greater effects of grazing on species composition on the less productive savannas. Other studies that have examined mammal effects on species composition at different levels of productivity have used rainfall as a productivity indicator (e.g. Chase et al. 2000). In our study, termite mounds have soils with higher nutrient content than savannas. Consequently, resource differences between the habitats are high. In the study area, herbivores preferentially forage on mound vegetation (Mobæk, Narmo & Moe 2005). Thus, to some extent, our treatments represented extremes, in the sense that termitaria have high community production and high herbivore intensity and savanna areas have low community production and low herbivore intensity. This pattern is consistent across African savanna areas with large, vegetated termitaria (Holdo & McDowell 2004; Loveridge & Moe 2004), where termitaria and savanna comprise different herbaceous species; termitaria are dominated by forbs and savanna areas by graminoids.
The influence of mammalian herbivores on herbaceous species composition increases with productivity up to a certain threshold (e.g. Bakker et al. 2006). Beyond this resource threshold, there may be changes in the relative dominance of major functional groups (i.e. forbs on termitaria and graminoids on savanna areas) not caused by large mammals, but by other factors, such as in our case, termites. Indeed, in their conceptual model, Hopcraft, Olff & Sinclair (2010) proposed that the availability of key environmental resources has profound consequences for herbivore regulation and ecosystem dynamics by simultaneously affecting multiple top-down and bottom-up processes. Termitaria are key resource areas in savannas and substantially influence spatial and temporal dynamics of soil nutrients, moisture and fire regimes (Konate et al. 1999; Sileshi et al. 2010).
Contrary to one of our predictions, we found that fencing gradually increased the ratio of graminoids to forbs. Fencing increased the total amount of graminoids but did not affect forbs (Table 2). Graminoids are generally regarded as a grazing-tolerant functional group because they recover quickly from defoliation and are believed to have co-evolved with the abundant and diverse ungulate fauna in the ecosystem (McNaughton 1984). In Wisconsin, USA, Rooney (2009) found that broadleaf herb cover was 63 times higher in exclusion plots and grass species were 2.2 times higher in plots grazed by white-tailed deer (Odocoileus virginianus). In our study, the relative cover of forbs and graminoids also changed with time, suggesting that factors other than grazing (e.g. temporal changes in rainfall) influence these cover changes.
Consistent with our final prediction, we found that some soil nutrients were significantly higher on termitaria compared with savanna areas. Levels of pH, Ca and Mg were higher on termitaria than on adjacent savanna. We found higher levels of Na in the adjacent savanna compared with termitaria, but N and C did not differ between the two habitats. The fact that N did not differ between mounds and savanna soil in our study does not necessarily mean that the fixation is the same in both habitats. Rather, plant nitrogen consumption is likely to be higher in termitaria habitats, which supports high densities of forbs (Moe, Mobæk & Narmo 2009). Forbs commonly gain a competitive advantage in nutrient-enriched habitats (Güsewell 2004). The dominance of nutrient-demanding grass species, such as C. dactylon, P. maximum and many forbs on termitaria may be related to their capacity to exploit nutrient-enriched termitaria. A similar pattern of increasing forb density on termite mounds and nitrogen-rich glades was reported for Pennisetum stramineum in Kenyan savanna (Riginos et al. 2009). Through physical disturbance of the soil and transport of nutrients vertically and horizontally, termites can create nutrient-rich areas that can support more herbaceous plant species (Glover, Trump & Wateridge 1964; Spain & McIvor 1988).
Grazing reduced the P content of soil, which increased after the exclusion of large herbivores. A similar increase in P following fencing was reported in the Serengeti National park in Tanzania by Anderson, Ritchie & McNaughton (2007) and in a study of grazing lawns in Nepal (Moe & Wegge 2008). It has been suggested that local-scale herbivore-induced losses of P may trigger increased P cycling rates from plant tissue consumption and turnover and P accumulation in animal biomass (particularly bones) that outstrip soil P mineralization (Chaneton, Lemcoff & Lavado 1996; Anderson, Ritchie & McNaughton 2007).
In this study, differences in herbaceous community composition were not related to large herbivores, but to termitaria and soil resources, indicating that mound building is the major mechanism by which termites induce strong spatial heterogeneity at the landscape scale. The influence of large termitaria on plant as well as animal species has been acknowledged by ecologists for decades (Burtt, Jackson & Potts 1942; Wild 1952) but has yet to be formally and explicitly applied in savanna ecology. Close evolutionary bonds that tie many plant species and mound-building termites together suggest that the co-evolved association developed over time may have important implications for savanna management. The conspicuous and wide distribution of large termitaria in the tropics and the fact that termitaria have a perpetual legacy lasting centuries through recolonization (Moore & Picker 1991) suggest a substantial co-evolutionary time-scale. Our results show that termite mounds facilitate the establishment of many herbaceous species, particularly forbs, not occurring in the adjacent grass savanna. Grazing by large mammalian herbivores contributes little to species variation in the herbaceous community. Consequently, it is important to account for the effect of important insect groups in addition to the more conspicuous large mammal guild in the studies of savanna vegetation dynamics.
The Uganda Wildlife Authority provided permission and allowed access to Lake Mburo National Park. The Norwegian Research Council and the Norwegian State Education Loan Fund funded the study. James Nkwatsibwe, Patrick Bodo and Rose Kentalu helped during fieldwork and Protase Rwaburindore assisted in plant collections and identification. Kari Klanderud and two anonymous referees provided constructive comments on an earlier draft.