Grazers, Browsers, And fire Influence The Extent And Spatial .

Transcription

Ecological Applications, 19(1), 2009, pp. 95–109Ó 2009 by the Ecological Society of AmericaGrazers, browsers, and fire influence the extent andspatial pattern of tree cover in the SerengetiRICARDO M. HOLDO,1,3 ROBERT D. HOLT,1ANDJOHN M. FRYXELL21Department of Zoology, University of Florida, Gainesville, Florida 32611 USADepartment of Integrative Biology, University of Guelph, Guelph, Ontario N1G 2W1 Canada2Abstract. Vertebrate herbivores and fire are known to be important drivers of vegetationdynamics in African savannas. It is of particular importance to understand how changes inherbivore population density, especially of elephants, and fire frequency will affect the amountof tree cover in savanna ecosystems, given the critical importance of tree cover for biodiversity,ecosystem function, and human welfare. We developed a spatially realistic simulation modelof vegetation, fire, and dominant herbivore dynamics, tailored to the Serengeti ecosystem ofeast Africa. The model includes key processes such as tree–grass competition, fire, andresource-based density dependence and adaptive movement by herbivores. We used the modelto project the ecosystem 100 years into the future from its present state under different fire,browsing (determined by elephant population density), and grazing (with and withoutwildebeest present) regimes. The model produced the following key results: (1) elephants andfire exert synergistic negative effects on woody cover; when grazers are excluded, the impact offire and the strength of the elephant–fire interaction increase; (2) at present populationdensities of 0.15 elephants/km2, the total amount of woody cover is predicted to remain stablein the absence of fire, but the mature tree population is predicted to decline regardless of thefire regime; without grazers present to mitigate the effects of fire, the size structure of the treepopulation will become dominated by seedlings and mature trees; (3) spatial heterogeneity intree cover varies unimodally with elephant population density; fire increases heterogeneity inthe presence of grazers and decreases it in their absence; (4) the marked rainfall gradient in theSerengeti directly affects the pattern of tree cover in the absence of fire; with fire, the woodycover is determined by the grazing patterns of the migratory wildebeest, which are partlyrainfall driven. Our results show that, in open migratory ecosystems such as the Serengeti,grazers can modulate the impact of fire and the strength of the interaction between fire andbrowsers by altering fuel loads and responding to the distribution of grass across thelandscape, and thus exert strong effects on spatial patterns of tree cover.Key words: Acacia; elephants (Loxodonta africana); GIS; migration; savanna dynamics; SD model;spatial coupling; spatial heterogeneity; Tanzania; tree–grass interactions; wildebeest (Connochaetestaurinus).INTRODUCTIONA long-standing challenge in ecology is the formulation of a general theory of tree–grass regulation insavannas (Belsky 1990, Scholes and Archer 1997,Higgins et al. 2000, Sankaran et al. 2004). Savannascan exhibit marked spatiotemporal variation in treebiomass, but it is still not well understood how thisvariation emerges from climate, edaphic factors, herbivores, fire, or interactions among these variables (Scholesand Walker 1993, Sankaran et al. 2004, 2005, Bond 2005,Bond et al. 2005, White 2006). In African savannas, it isclear that three main factors can strongly affect therelative biomass of woody and herbaceous vegetation:soil moisture, fire, and mammalian herbivory (Walker1987, Scholes and Walker 1993, Bond 2005). What is lessclear is how these factors interact dynamically, bothlocally and across productivity gradients, to structurevegetation, given that strong feedbacks can occur amongplants, herbivores, and fire (Frost and Robertson 1987).Such feedbacks have the potential to greatly influencesystem behavior, and it is important to understand suchfeedbacks to develop long-term management strategiesShifts in the amount of woody cover as a result ofclimate change, fire, herbivory, and human agency havethe potential to exert strong impacts on ecosystemfunction in savannas (Ringrose et al. 1998, Hoffmannet al. 2002, Jackson et al. 2002, Scanlon et al. 2005,Pringle et al. 2007). Savannas comprise 40% of theterrestrial land mass and sustain a significant fraction ofthe population of many developing countries (Scholesand Walker 1993), so these changes may have deepimpacts on human welfare globally. It is thus critical thatwe obtain a better quantitative understanding of how themultiple factors that influence savanna woody coverinteract with each other, both for the management ofparticular ecosystems and to assess the regional andglobal implications of shifts in shrub and tree cover.Manuscript received 28 November 2007; revised 3 April2008; accepted 6 May 2008. Corresponding Editor: N. T.Hobbs.3 E-mail: rholdo@ufl.edu95

96RICARDO M. HOLDO ET AL.and to project the impacts of climate change and pressurefrom a rising human population.Further complexity is added by the fact that Africansavannas are often characterized by the presence of twodistinct major vertebrate herbivore guilds: grazers, whichfeed primarily on grasses and forbs, and browsers, whichfeed primarily on woody vegetation. Few studies (e.g., vanLangevelde et al. 2003) have so far explored the relativeimportance of grazers and browsers (or for that matter,mixed feeders) as determinants of tree-to-grass ratios insavannas, and to the best of our knowledge, none have sofar explicitly investigated how grazing, browsing, and fireinteract to determine savanna tree cover. Grazers andbrowsers can both modulate the effects of fire on treedynamics by affecting fuel biomass, either directlythrough grass consumption, or indirectly by reducingthe competitive effect of trees on grasses (Norton-Griffiths1979, Frost and Robertson 1987, Holdo 2007). In adynamic system, grazers and browsers may also potentially interact with each other by exerting reciprocal effectson their own food resources. This three-way browser–grazer–fire interaction may be particularly important inAfrican savannas because these systems are oftendominated by megafaunal assemblages that fill both thegrazing and browsing niches (Owen-Smith 1988).Understanding the interactions and feedbacks amonggrazers, browsers, and fire may be critically important forthe management of woody cover in many savannaecosystems. Managers have long been concerned bothby unwanted increases (bush encroachment in rangelands)and losses (declines in woody canopy cover in protectedareas) in woody cover (Laws 1970, Martin et al. 1992,Prins and Vanderjeugd 1993, Smit and Rethman 2000,Western 2006), so the development of predictive models ofwoody biomass dynamics is an important goal for thesesystems. Given the importance of tree cover for climate,ecosystem processes, biodiversity, and human livelihoodswithin the savanna biome (Scholes and Walker 1993,Hoffmann et al. 2002, Jackson et al. 2002), it is importantto understand how the management of fire and herbivoreabundance might impact tree dynamics, since they are tosome extent under human control.We are here primarily concerned with the potential forlosses in tree cover in savanna systems that are both fireprone and dominated by elephants. Elephants standapart from most other browsers because of their abilitynot only to slow the recruitment of trees from small tolarge size classes, but to rapidly reduce tree cover overshort periods of time by toppling and ringbarkingmature trees (Laws 1970, Guy 1989, Ruess and Halter1990, Holdo 2006). Across Africa, elephants have oftenbeen implicated as the key drivers of large changes intree abundance (Buechner and Dawkins 1961, Laws1970, Dublin et al. 1990, Ben-Shahar 1993, Van deVijver et al. 1999, Western and Maitumo 2004). Weaddress the question of tree cover regulation in savannasby developing a semimechanistic simulation model(dubbed SD, for Savanna Dynamics) to investigate theEcological ApplicationsVol. 19, No. 1role of browsers, grazers, and fire in determining theamount of tree cover in savannas. We focus on a few keydrivers and system components within a framework thatallows herbivores to move adaptively across thelandscape in response to changes in food availability(McNaughton 1985, Fryxell et al. 2004) brought aboutby fire, consumption levels, and rainfall. This model thusenables us to explore the combined effects of fire,browsing, and grazing on vegetation structure.SD differs from previous simulations of fire and/orherbivore effects in savannas (e.g., Starfield et al. 1993,Baxter and Getz 2005, Holdo 2007, Liedloff and Cook2007) in a number of key respects: (1) unlike previousmodels, we use a spatially realistic framework tosimulate vegetation dynamics at the landscape scale,using a GIS-based lattice that incorporates realisticspatial variation in edaphic factors, as well as nutrientand climatic gradients; (2) we take into account theeffects of herbivory on both grasses and trees; and (3) weincorporate feedbacks of the vegetation on herbivoresand fire. Thus our model treats herbivory and fire asdynamic processes rather than only as fixed drivers. Webelieve that our model can thus be potentially appliedacross a wide range of conditions, permitting assessmentof a wide range of management and conservationscenarios in African savannas. SD is a model ofintermediate complexity, with less mechanistic detailthan models such as SAVANNA (Boone et al. 2002) andFLAMES (Liedloff and Cook 2007), but with sufficientcomplexity to generate realistic landscape-level simulations of vegetation, fire, and herbivore dynamics, unlikethe case for simpler models (van Langevelde et al. 2003,D’Odorico et al. 2006).We use the Serengeti ecosystem of East Africa as acase study for the investigation of vegetation–herbivore–fire dynamics in the present paper, for a number ofreasons: first, it is an open, fire-prone ecosystemcharacterized by strong abiotic (both climatic andedaphic) gradients and by the presence of both grazingand browsing ungulate herbivore guilds that canprofoundly affect vegetation structure (Sinclair 1975,Pellew 1983, McNaughton 1985). The heterogeneity inabiotic and biotic factors across the landscape meansthat the Serengeti essentially incorporates the range ofvariables that determine vegetation structure acrossmost African savannas, and thus this system representsan excellent laboratory for the study of vegetationdynamics in savannas. Second, historic records revealstrong shifts in tree-to-grass ratios in the Serengeti as aresult of changes in the fire regime and herbivorepopulations (Norton-Griffiths 1979, Sinclair 1979), andsuch records provide a rich source of data forparameterizing and validating models. Finally, theSerengeti is itself of great importance for the conservation of biodiversity and ecological processes in thesavanna biome (Sinclair et al. 2007), and we hope toprovide a useful management tool for this and othersavanna ecosystems.

January 2009HERBIVORES, FIRE, AND TREE COVER97FIG. 1. (a) Isohyets (mean mm per annum between 1960 and 2001); (b) grass N concentration (%); (c) plains and woodlandhabitats; and (d) 1972 percent canopy cover in the greater Serengeti ecosystem (GSE). The Serengeti National Park boundary isshown in outline.After introducing the model structure, we show thatthe model provides good fits to the long-term dynamicsof key variables in the Serengeti. We then conductsimulations to explore the future trajectory of the systemand the sensitivity of the model to uncertainty in theparameters. Our primary objective in this paper is toexamine how interactions between fire, browsing, andgrazing determine vegetation structure across rainfalland fertility gradients. We first address the question ofhow contrasting fire frequencies and elephant population densities are predicted to determine the averagefuture trajectory of tree cover in the Serengeti-Maraecosystem, and how the elephant–fire interaction is inturn affected by the presence of grazers. We theninvestigate how these factors are predicted to affectpatterns of tree cover across space. Finally, we examinethe effects of elephant population density and fire on thewildebeest population and the effect of the grazer–browser interaction on fire.MATERIALANDMETHODSStudy systemThe greater Serengeti ecosystem (GSE hereafter)comprises .30 000 km2 of savannas and grasslands inTanzania and Kenya. The GSE includes two nationalparks, several game reserves and game managementareas, and unprotected land along its periphery, and isthus subject to a range of anthropogenic resourceutilization regimes. Following well-established precedent, we define the GSE as the area that approximatelybounds the resident and migratory wildebeest populations of the Serengeti-Mara complex (Maddock 1979,Sinclair 1979). We presently ignore the effects of humanactivity both within the protected areas and in the smallfraction of the western GSE that is settled. We willaddress these effects and socio-ecological interactions inan upcoming paper.A marked southeast to northwest rainfall gradientcharacterizes the system (Fig. 1a), as well as a fertilitygradient that runs approximately opposite to the rainfallgradient (Fig. 1b). The ecosystem is predominantlywoodland savanna, but significant areas of puregrassland occur, notably in the southeastern plains andthe Mara in the northern sector (Fig. 1c). Throughoutmuch of these grassland areas, trees are almost entirelyabsent due to the presence of a hardpan layer close tothe soil surface (Belsky 1990). In the woodland habitat,on the other hand, the amount of tree cover can varyconsiderably (Fig. 1d). The rainfall gradient is the enginethat drives the seasonal migration of the wildebeest and

98RICARDO M. HOLDO ET AL.FIG. 2. Schematic representation of the key players in theSerengeti plant–herbivore dynamics (SD) model and theirinteractions. The dashed line indicates a weak effect ofelephants on grasses.other ungulates (Sinclair 1979, Boone et al. 2006).During the wet season, when grass production in theSerengeti plains is high, the wildebeest migrate south,returning to the northern woodlands as green grassbecomes increasingly restricted to areas with dry-seasonrainfall (McNaughton 1979).Model descriptionThe savanna dynamics (SD) model focuses on anumber of key processes in the ecosystem (see Fig. 2):grass and tree growth, mortality, and consumption;herbivore population dynamics and movement; and firedynamics; all of which are influenced by rainfall. Themodel is implemented on a spatially explicit, GIS-basedframework, which facilitates comparison of the modelwith empirical data sets for particular systems (in thiscase the Serengeti, but SD may easily be modified andapplied to other ecosystems). The state variables in themodel are tracked within cells embedded in a lattice. Thelattice represents the GSE, and covers 30 700 km2,divided into 307 10 3 10 km cells. The choice ofboundaries for the GSE and cell size follows Maddock(1979). The agents represented by the model (Fig. 2)include ‘‘keystone’’ species (two dominant herbivores,wildebeest Connochaetes taurinus, which as noted above‘‘define’’ the Serengeti by their migratory behavior, andelephants Loxodonta africana), as well as the two majorplant physiognomic categories, trees and grasses. Themodel distinguishes green from dry-grass biomass, butdoes not track separate grass species. Furthermore,because fire-driven mortality is strongly dependent ontree size (Pellew 1983), the model tracks the sizestructure of trees. We model single generic grass andtree species. (For the latter we use Acacia tortilis,because data are readily available for this species and itis the most abundant and widely distributed tree speciesin the ecosystem.)Wildebeest are the dominant grazers in the Serengeti(Sinclair 2003), accounting (by our estimate) for overEcological ApplicationsVol. 19, No. 1half the herbaceous biomass consumed by largeherbivores in this ecosystem. Elephants are mixedfeeders whose impact is most readily apparent on thetree community (Croze 1974a, b, Pellew 1983, Dublin etal. 1990). Although giraffe may affect tree growth in theSerengeti (Pellew 1983), our primary concern in thispaper is with the role of the elephant population, whichhas been expanding rapidly over the past decade. In thepresent version of the model, we do incorporate giraffebrowsing indirectly (in the tree growth coefficients) buttreat it as a constant. Future extensions of the model willinclude other species (e.g., resident herbivores such asbuffalo, giraffe, and carnivores) as explicit dynamicalvariables.The model is time discrete and uses different timesteps for different compartments, to reflect a balancebetween crucial biological detail and computationalefficiency. Potentially rapid changes in grass biomassover short time periods (McNaughton 1985) dictate thatgrass growth, consumption, and decay occur on a dailytime scale. Wildebeest herbivory and local populationdynamics are also modeled with a daily interval tomatch the rapid dynamics of their resource. Wildebeestmovement among cells occurs on a weekly scale (forfaster computational execution; using this longer timescale for movement does not affect model results). Treedynamics, by contrast, follow an annual time step.Rainfall varies monthly. During simulations, a years’worth of rainfall data (12 months from November of theprevious year to October, assuming that the wet seasonbegins in early November) is randomly selected from the1960–2006 historical record and used to drive grass andtree growth for one annual cycle. This process isrepeated for each year of the simulation. This climaticdriver and the fire submodel are the only stochasticcomponents in an otherwise deterministic model.The model landscape.—The model uses three types ofraster data sets or maps generated through a GISanalysis: rainfall, habitat type, and plant nitrogencontent (Fig. 1). These GIS layers play a dual role. Inaddition to being used to fit free parameters in themodel, they are also used to generate model inputs forsimulations. We created monthly rainfall layers for theGSE for the period 1960–2006 using monthly rainfalldata from 204 gauges distributed throughout theSerengeti ecosystem (TAWIRI records). We generatedthe rainfall raster files with an inverse distance-weightedtechnique in ArcGIS 9.1 (ESRI, Redlands, California,USA) using 12 nearest neighbors and power 2 (Legendreand Legendre 1998:747–748). We produced the habitattype map by joining shapefile and raster vegetationlayers from multiple sources into a composite layer forthe entire GSE (Oindo et al. 2003; M. Coughenour,unpublished data; D. Herlocker, unpublished data; K. L.Metzger, unpublished data). We used multiple sourcesbecause no single map currently available covers theentire GSE. The rationale for developing the habitatmap was to identify treeless areas of the landscape to

January 2009HERBIVORES, FIRE, AND TREE COVER99FIG. 3. (a) Grass production as a function of annual rainfall predicted by McNaughton’s (1985) empirical function and thecorresponding savanna dynamics (SD) model fit; (b) model fit to wildebeest census data; (c) model fit to actual and simulatedmonthly locations of ‘‘center of mass’’ of Serengeti wildebeest population averaged over the 1969–1972 time period; (d) model fit tofire extent data over a 40-year period in the Serengeti.prevent the model from allowing tree growth in habitatswhere woody vegetation is excluded due to edaphicconstraints (Sinclair 1979, Belsky 1990). In addition, forthe woodland habitat, where trees do occur, we wereable to infer past patterns of tree canopy cover frommaps developed by M. Norton-Griffiths (unpublisheddata) based on aerial photography. We used these mapsto initialize our model runs and as a data layer for thefitting of our wildebeest movement submodel (R. M.Holdo, R. D. Holt, and J. M. Fryxell, unpublish

Serengeti is itself of great importance for the conserva-tion of biodiversity and ecological processes in the savanna biome (Sinclair et al. 2007), and we hope to provide a useful management tool for this and other savanna ecosystems. 96 RICARDO M. HOLDO ET AL. Ecological Applications Vol. 19, No. 1