Limits to exploitation: dynamic food web models predict the impact of livestock grazing on Ethiopian wolves Canis simensis and their prey


  • Flavie Vial,

    Corresponding author
    1. Boyd Orr Centre for Population and Ecosystem Health, Institute of Biodiversity, Animal Health and Comparative Medicine, College of Medical, Veterinary and Life Sciences, University of Glasgow, Glasgow G12 8QQ, UK
    2. Wildlife Conservation Research Unit, The Recanati-Kaplan Centre, Zoology Department, University of Oxford, Tubney House, Tubney OX13 5QL, UK
      Correspondence author. Department of Infectious Disease Epidemiology, Imperial College London, St Mary’s Campus, London W2 1PG, UK. E-mail:
    Search for more papers by this author
  • David W. Macdonald,

    1. Wildlife Conservation Research Unit, The Recanati-Kaplan Centre, Zoology Department, University of Oxford, Tubney House, Tubney OX13 5QL, UK
    Search for more papers by this author
  • Daniel T. Haydon

    1. Boyd Orr Centre for Population and Ecosystem Health, Institute of Biodiversity, Animal Health and Comparative Medicine, College of Medical, Veterinary and Life Sciences, University of Glasgow, Glasgow G12 8QQ, UK
    Search for more papers by this author

Correspondence author. Department of Infectious Disease Epidemiology, Imperial College London, St Mary’s Campus, London W2 1PG, UK. E-mail:


1. Anticipating the consequences of, and sustainable limits to, human exploitation of ecosystem resources requires a quantitative understanding of food web dynamics. While dynamic food web models are commonly used to investigate the impact of human exploitation on marine ecosystem structure and dynamics, they are less commonly applied to human-induced disturbances in terrestrial food webs.

2. The intensifying and expanding use of rangelands for domestic livestock production creates a potential conflict with the conservation of extensive populations of wildlife. The Bale Mountains National Park (BMNP) is an example of an area that increasingly has been used as an open-access resource by pastoralists and their herds. Livestock grazing, through its impact on herbivorous rodents, is suspected to negatively affect the critically endangered Ethiopian wolf Canis simensis, a rodent specialist.

3. We developed a series of simple dynamic food chain models to explore the interactions between these trophic levels and how they might be affected by livestock grazing. We also explored how predictions made about these trophic dynamics are affected by the type of functional response linking the different trophic levels.

4. If rodent and/or wolf populations exhibit extreme type 2 responses to resource availability, they can both remain at densities close to their ungrazed equilibrium as livestock density increases, but will rapidly crash once vegetation biomass collapses as a result of the increased grazing pressure. The model predicts the maximum sustainable livestock density to lie between 32 and 117 TLU·km−2, above which populations of wolves are expected to become locally extinct.

5.Synthesis and applications. We show that dynamic trophic modelling is an informative approach when investigating the response of terrestrial ecosystems to human-induced disturbance. In particular, the models described in this paper provide a first step towards enabling managers to predict the implications of changing patterns of human impact on rangelands and support the identification of sustainable grazing grounds for livestock. The models also reveal that monitoring primary productivity alone may be a simple and effective way to detect stresses on food chains and predict the impact of increased livestock grazing on higher trophic levels. In the context of the BMNP, the models predict that some areas of high conservation value appear to be exploited unsustainably, thereby possibly affecting long-term conservation goals for the Park’s flagship species.


As human activities play an increasingly important role in affecting ecosystem processes, the ability to predict the direct and indirect effects of these impacts on these processes becomes a priority. For example, a better understanding of how competition between native species and domestic species affect trophic interactions and other key ecosystem processes is crucial to assessing the sustainability of current human land-use patterns and informing the development of relevant management strategies. Food webs provide insight into the trophic structure and energy flows of ecosystems (Vander Zanden, Casselman & Rasmussen 1999; Cohen, Jonsson & Carpenter 2003), the factors affecting trophic dynamics, as well as the trophic effects of disturbances (Jenkins, Kitching & Pimm 1992; Townsend et al. 1998). Dynamic trophic modelling can be informative when exploring the ecological issue of biomass partitioning between competing species, for example (Fox 2005), as well as the more general impact of human-induced changes in trophic relationships on an ecosystem’s structure and dynamics (Walters, Christensen & Pauly 1997).

There are significant associations between human population density and biodiversity hotspots in Africa (Balmford et al. 2001; Sachs et al. 2009). At these sites, conflicts over natural resources are frequent (Stewart 2002) and often centred on contested access to land (Peluso 1993). An example of this human–wildlife conflict is being played out in and around Ethiopia’s Bale Mountains National Park (BMNP) (06°41′N, 39°03′E and 07°18′N, 40°00′E). The Bale mountains represent the largest area of afroalpine habitat in Africa (Yalden 1983), and harbour a diverse array of endemic and range-restricted species, including the largest remaining populations of Ethiopian wolves Canis simensis Rüppell, 1840. The Ethiopian wolf, with a total population of <500 individuals, is the rarest canid in the world. They are at the top of one of the most simple, yet critical, food chains in afroalpine pastures (Sillero-Zubiri, Tattersall & Macdonald 1995), preying upon the rodent fauna, especially the endemic giant mole rat Tachyoryctes macrocephalus and two species of murine rodents, Arvicanthis blicki and Lophuromys melanonyx. These species combined represent an estimated 90% of the wolves’ diet (by volume) (Sillero-Zubiri & Gottelli 1995), and feed primarily on the above-ground parts of grasses and flowering plants (Alchemilla spp. in particular) on the afroalpine pastures (Yalden 1988).

Core wolf ranges in the BMNP are located in the Web valley, Morebawa and on the Sanetti plateau. Ethiopian wolf populations have been affected by diseases transmitted from domestic dog populations (Randall et al. 2006) but livestock grazing, through its impact on the wolves’ rodent prey, is considered to represent a more profound long-term threat to their persistence (Stephens et al. 2001; Vial 2010). Livestock density estimates for 2007/2008 based on distance-sampling along 500 km of transects (Vial 2010) showed densities of cattle and caprines (sheep and goats) to be c. 195 tropical livestock units (TLU) km−2 (95% CI: 125–325 TLU·km−2) in the Web valley, c. 149 TLU·km−2 (95% CI: 69–342 TLU·km−2) for Morebawa and 49 TLU·km−2 (95% CI: 13–194 TLU km−2) for the Sanetti plateau in 2007–2008 [1 TLU = 1·5 cattle, 11 caprines or 1·5 equid (Boudet & Riviere 1968)]. It has been shown that rodent biomass declines as livestock density increases along a grazing gradient and that rodent density increases in response to the experimental removal of livestock grazing using exclosures (Vial 2010). Such experiments have also revealed some evidence that the impact of livestock on rodent biomass is concurrent with changes in the vegetation. In particular, the removal of livestock has a positive impact on the biomass of Alchemilla spp. and grasses, and results in smaller home ranges and higher reproductive success for L. melanonyx, an indication that resources may be more limited for some rodent species on grazed sites (Vial 2010).

We apply classical and modified Lotka-Volterra (LV) predator–prey models to explore the potential effects of exploitative competition between livestock and rodents for primary production in BMNP and its possible repercussions on trophic levels higher in the food chain. Sustainability can then be evaluated in terms of the food chain’s ability to withstand disturbances (Holling 1973), that is the capacity of a system to absorb change while preserving its structure and dynamics. There are two common ways of measuring this property: stability and resilience. Here, we will use stability as ‘the propensity of a system to attain an equilibrium condition’ (Holling 1986) (quantified as the coefficient of variation around mean equilibrium biomasses), while we will define resilience as the ‘domain (quantified here as a livestock stocking rate) over which disturbance can be experienced’ while still retaining an equilibrium condition (sensuHolling 1973). These models provide a framework for the objective examination of biomass flows and standing crops, an approach rarely adopted when considering the following ecological questions: What is the impact of increasing competition (in our case livestock competing with rodents) on the food chain’s stability and resilience? Does this impact depend on the type of functional response that links the different trophic levels?

More specifically for our case study, we also ask whether livestock densities in the Web valley, Morebawa and the Sanetti plateau exceed our models’ estimated resilience to disturbance thereby threatening Ethiopian wolves’ persistence in part of their range in BMNP?

Materials and methods

Model structure

We explore the dynamics of a three-trophic level model where biomass (kg) per unit area (km−2) of vegetation (Xv), rodents (Xr) and wolves (Xw) is represented by a set of ordinary differential equations:

image(eqn 1)
image(eqn 2)
image(eqn 3)

The parameters bi are the rates of change of biomass in the absence of other trophic levels (kg·km−2·year−1) and therefore positive for vegetation, and negative for rodents and wolves. The aii terms govern the density dependence at each trophic level; L is the biomass of livestock (kg·km−2) and is assumed to be imposed on the landscape and not subject directly to the trophic dynamics so is represented as a parameter; κij is the conversion efficiency of biomass from the jth level to the ith.

Ci describes the functional response that relates biomass consumption per kg of consumer per year to resource availability. In this model, there are three such functional responses: Cr(Xv) – the biomass of vegetation consumed per kg of rodent as a function of the biomass of vegetation; CL(Xv) – the biomass of vegetation consumed per kg of livestock as a function of the biomass of vegetation; and Cw(Xr) – the biomass of rodents consumed per kg of wolf as a function of the biomass of rodents. We investigate combinations of these functional responses, assuming them to be either type 1 (linear) or an extreme version of type 2, where per capita consumption is independent of resource availability so long as resource is available at all.


We parameterize functional responses by considering an equilibrium state of the system in which the effects of livestock grazing is negligible (denoted by the dash), and in which wolves (inline image) are present at a biomass density of 30 kg·km−2 [a wolf weighs ∼14 kg, and are often observed at densities of 2–3 wolves per km−2 (Sillero-Zubiri 1994)], rodents (inline image), a mixture comprising murine rodents and giant mole rats, present at a biomass density of 2860 kg·km−2; and vegetation (inline image) at 385 000 kg·km−2 (Woldu 1986; Mwendera, Saleem & Woldu 1997; Asefa et al. 2003; Yayneshet, Eik & Moe 2003). Wolves are assumed to consume an equivalent biomass of one giant mole rat (0·62 kg) per day (Sillero-Zubiri & Gottelli 1995), hence inline image = (0·62/14) × 365 = 16·16 kg·km−2·year−1. Rodents are assumed to consume a biomass of vegetation equivalent to 20% of their own biomass per day (Grodziński & French 1983), hence inline image = 0·2 × 365 = 73 kg·km−2·year−1.

With type 1 functional responses, as assumed by classical LV dynamics, Cj(Xi) = aijXi, and thus inline image. In the case of the extreme type 2 functional response, consumption is independent of resource availability and fixed at inline image. In this case, it is assumed individuals compensate for resource scarcity through an increased effort in acquiring resources. Obviously there are limits to such compensation, but the intention here is to encompass reality; it is likely that the real functional responses will lie someway between these two extremes.

The trophic efficiency parameters κrv and κwr are assumed to be 5% and 10% respectively (Colinvaux & Barnett 1979; Grodziński & French 1983). The bi terms are estimated assuming a doubling time for vegetation (at low densities) of 1 month, and a half-life of rodents and wolves of 1 month and 6 months respectively. The aii terms are then the only remaining unknown parameters and can be solved for at the system equilibrium point, setting the right hand-sides of eqns 1–3 equal to zero.

Livestock grazing pressure is modelled as TLUs each equivalent to 250 kg and requiring 2300 kg·year−1 of dry vegetation biomass. Thus, assuming a wet-dry conversion factor of 2·2 (Le Houérou & Hoste 1977), inline image = 2300 × 2·2/250 = 20·44 kg·year−1. We examine how the system responds to a range of grazing pressures equivalent to between 0 and 200 TLU·km−2.

Table 1 provides a summary of the parameters used in the trophic models as well as references for the empirical work they are based on.

Table 1.   Summary of trophic model parameters: point estimates for the instantaneous growth rates (bi) and per capita species effects (aij) as used in the trophic models
  1. Empirical work and data on which parameter estimates are based are referenced as follow: (a) Asefa et al. (2003); (b) Colinvaux & Barnett (1979); (c) Grodziński & French (1983); (d) Le Houérou & Hoste (1977); (e) Mwendera, Saleem & Woldu (1997); (f) Rubal, Haim & Choshniak (1995); (g) Sillero-Zubiri (1994); (h) Sillero-Zubiri & Macdonald (1997); (i) Vial (2010); (j) Woldu (1986); (k) Yayneshet, Eik & Moe (2003).

awv (km2·kg−1·year−1)0 
avw (km2·kg−1·year−1)0 
arl (km2·kg−1·year−1)0 
arr (km2·kg−1·year−1)−1·7 × 10−4 
arv (km2·kg−1·year−1)9·5 × 10−6a, c, e, f, j and k
avr (km2·kg−1·year−1)−1·9 × 10−4a, e, f, j and k
aww (km2·kg−1·year−1)−7·5 × 10−3 
awr (km2·kg−1·year−1)5·6 × 10−4a and g
avl (km2·kg−1·year−1)−5·3 × 10−5a, d, e, j and k
avv (km2·kg−1·year−1)−1·9 × 10−5 
arw (km2·kg−1·year−1)−5·6 × 10−3g
bv (kg·km−2·year−1)7·82a, e, j and k
br (kg·km−2·year−1)−3·00i
bw (kg·km−2·year−1)−1·39h

Model analysis

We set out to examine how grazing pressure changes the biomass of vegetation, rodents and wolves at equilibrium. Of particular interest is the grazing pressure (Lmax) at which wolf biomass can no longer be sustained. We also examine the stability of this equilibrium point.

Model 1 (M1) assumes that all functional responses are type 1, and takes the simple LV form. The equilibrium points inline image can be recovered by inversion of the interaction matrix A (containing the aij elements) and multiplication by the vector b (containing the bi elements): inline image. We also evaluated the local stability of these equilibrium points (conditional on their feasibility: inline image for all i) by examining the dominant eigenvalue of the appropriately formulated Jacobian matrix (J) evaluated at inline image.

We examine four model variants:

  •  M2: Similar to M1 except that the impact of livestock grazing on vegetation is modelled according to a type 2 response.
  •  M3: Similar to M2 except that the impact of rodent grazing on vegetation is also modelled according to a type 2 response.
  •  M4: Similar to M2 except that the impact of wolf predation on rodents is modelled according to a type 2 response.
  •  M5: All functional responses are type 2.

With the introduction of alternative non-type 1 functional responses, the model loses its classical LV form and the equilibrium is most easily evaluated numerically. Sensitivity and elasticity analyses were performed on the parameters about which there was the greatest uncertainty: bw, br, bv and inline image.


What is the impact of increasing disturbance by livestock on the food chain’s stability and resilience? Does this impact depend on the functional response that links the different trophic levels?

Equilibrium biomasses of the different trophic levels respond differently to increased disturbance by livestock depending on the type of functional responses (Fig. 1). Equilibrium biomass of vegetation declines steadily with increasing grazing pressure, but more quickly when rodents maintained a type 2 response to vegetation (models M3 and M5; Fig. 1a). If rodents and wolves exhibit a type 1 response (as in M1 and M2), both populations steadily decline as a result of increasing livestock grazing pressure on the vegetation (regardless of the type of functional response exhibited by livestock). If rodents exhibit a type 2 response (as in M3 and M5), the consumer species remain unaffected by increasing livestock grazing pressure as populations of rodents and wolves stay close to their initial equilibrium until the vegetation biomass crashes. If wolves exhibit a type 2 trophic response (as in M4), the predators remain largely unaffected by increasing livestock grazing pressure with densities close to their initial equilibrium until the rodent population crashes (Fig. 1b,c).

Figure 1.

 Changes in the equilibrium biomasses of vegetation (a), rodents (b) and Ethiopian wolves (c) as livestock density increases under all five model scenarios.

The analyses of all five models indicates that the equilibrium is locally stable whenever it was feasible (i.e. biomass at all trophic levels >0), but depending on the functional response linking the trophic levels, increasing livestock grazing pressure could push the system into a less locally stable state (Fig. 2). The system’s resilience, Lmax, for the LV model (M1) is estimated at 41 TLU·km−2 (Table 2). By numerically integrating the differential equations (eqns 1–3), we can also obtain estimates of the system’s resilience, Lmax, for the modified LV models 2–5 (Table 2). Maximum resilience to disturbance by livestock (117 TLU·km−2) was achieved under scenarios where rodents exhibited a type 2 response to vegetation availability, regardless of the wolves’ functional response to rodent availability. Minimum resilience occurred when wolves exhibited a type 2 response but their prey a type 1 response to vegetation availability. The probable true functional response of the consumer species in our trophic chain lies somewhere inbetween those two extreme scenarios and so we expect the system’s true resilience to lie between 32 and 117 TLU·km−2.

Figure 2.

 Changes in the dominant eigenvalue (λmax) of the Jacobian matrix governing the equilibrium point (in which wolves are represented) for each of the five trophic models as livestock density on the afroalpine pastures increases. Models 1 and 2 both predict that the stability of the equilibrium point is steadily reduced as grazing pressure increases. Models 3 and 5 predict no change in the stability of the equilibrium point until Lmax is reached and then the food chain suddenly collapses. Model 4 exhibits intermediate behaviour. The arrows indicate the estimated average livestock densities in the three wolf sub-populations’ ranges in 2007/2008 (from left to right): the Sanetti plateau, Morebawa and the Web valley.

Table 2.   Summary of five trophic models with the type of trophic response assumed for each consumer species
ModelFunctional responsesLmax
  1. The resilience of the food chain (Lmax) represents the estimated maximum possible sustainable livestock density parameters before wolves are eliminated from this food chain.

M1Type 1Type 1Type 140 TLU·km−2
M2Type 2Type 1Type 138 TLU·km−2
M3Type 2Type 2Type 1117 TLU·km−2
M4Type 2Type 1Type 232 TLU·km−2
M5Type 2Type 2Type 2116 TLU·km−2

Elasticity analyses revealed that uncertainty in the estimate of the wolves’ instantaneous growth rate (bw) resulted in the biggest difference in the estimate of the system’s resilience under scenario M1. If bw is overestimated by 5%, the resulting Lmax will be underestimated by 12%. Similarly, a 5% change in the estimate of br resulted in >5% difference in Lmax while bv and inline image were found to have little effect on the resilience of the system. We performed sensitivity analyses on all four parameters under both extreme scenarios M1 and M5 (Fig. 3). In M1, an increase in the instantaneous growth rates (for bw and br this means becoming more negative) results in an increased Lmax (Fig. 3a, c and e), with the effect being more pronounced (contours are tighter) for bw (Fig. 3a). However, under M5, the system’s resilience is less sensitive to changes bw and br (Fig. 3b,d) although biologically plausible food chains did not exist for some combinations of parameters (horizontal zero contour). An increase in vegetation equilibrium biomass inline image (Fig. 3e,f) has a positive impact on Lmax under both M1 and M5 (tighter contours suggest even more so for the latter).

Figure 3.

 Contour plots showing the predicted Lmax under the extreme scenarios M1 (left – all type 1 trophic responses) and M5 (right – all type 2 trophic responses) as a function of bw and inline image, br and inline image and bv and inline image. The white circles indicate the combination of parameter estimates used in the trophic models.


Do livestock densities in BMNP threaten Ethiopian wolves’ persistence?

We have used a simplified representation of the food chain supporting the largest remaining population of Ethiopian wolves to explore the sensitivity of trophic dynamics to increased levels of grazing pressure by domestic livestock. Depending on assumptions about the shapes of the various functional responses, the models indicate that the system’s resilience Lmax (the maximum density of livestock that is consistent with the maintenance of even a single pack of wolves), lies between 32 and 117 TLU·km−2. The Web valley and Morebawa are currently stocked at an estimated 195 and 149 TLU·km−2 respectively, and therefore lie outside of the models projected sustainable range. The intensity of grazing pressure is illustrated by the fact that the model estimates the ratio of primary productivity consumed by livestock relative to that consumed by rodents to be about 5 : 1 for the Web valley for the current stocking rate. The model outputs suggest that both systems, in their current state, may already be unstable while the Sanetti plateau (49 TLU·km−2) currently lies within our resilience ‘envelope’ for the system. The results of the sensitivity analyses suggest that, under the M1 scenario, both the Web valley and Morebawa would still be overstocked for inline image = 840 000 kg·km−2, the maximum Ethiopian afroalpine vegetation biomass density empirically recorded (Woldu 1986). However, under the M5 scenario, current livestock densities could be sustainable in both areas for inline image ≥ 617 000 kg·km−2.

Our resilience estimates are consistent with estimates of the maximum sustainable livestock stocking rates for all three ranges using a simpler static model developed by Le Houérou & Hoste (1977). Their model, linking primary productivity to rainfall in African rangelands, and applied to the Web valley, Morebawa and the Sanetti plateau predicts a maximum livestock stocking rate of 44, 43 and 34 TLU·km−2 for sustainable livestock production respectively.

Methodological considerations: model and assumptions

In the absence of more detailed data we assumed primary productivity is constant across all three wolf ranges. However, the shrublands on Sanetti plateau are generally considered to be a less productive system (Miehe & Miehe 1993) and therefore it is likely that Lmax for Sanetti is lower than the one predicted by this trophic model, and the food web may be closer to an unsustainable condition than we predict.

Knowledge of functional responses governing trophic interactions is critically important to applied aspects of consumer–resource biology. Specialist predators, such as the Ethiopian wolf, tend to exhibit a type 2 functional response with a curve rising rapidly as prey density increases (Andersson & Erlinge 1977), even when these predators are solitary hunters, as for example the Eurasian lynx Lynx lynx (Nilsen et al. 2009). However, all three major types of functional responses have been observed in herbivorous small mammals [sometimes within the same species (Klinger & Rejmánek 2009)]: from a linear intake rate (type 1) in brown lemmings Lemmus sibericus (Batzli, Jung & Guntenspergen 1981), to type 2 and type 3 functional responses in white-footed mice Peromyscus leucopus preying on moth pupae or on sunflower seeds respectively (Elkinton, Liebhold & Muzika 2004). As the three main rodent prey species were combined into one rodent trophic ‘class’ to simplify the description of the food chain dynamics, this model only allows us to look at the response of the trophic ‘class’ rather than the response of individual species to changes in resource availability. We consider that a type 3 rodent functional response is unlikely to arise in our model given that all the different plant species have been combined together into a single vegetation class. While L. melanonyx has been documented to incorporate invertebrates in its diet, it remains primarily an herbivorous species (Yalden 1988).

Similarly, the predictions of the model are based on the assumption that Ethiopian wolves’ diet is fixed and consists exclusively of rodents. In Bale, the contribution of different species of prey to the diet of the Ethiopian wolf correlates with their abundance (Sillero-Zubiri & Gottelli 1995). Circumstantial evidence also suggests that wolves may prey more frequently on livestock, or become crepuscular or nocturnal when human interference is severe in densely populated areas (Marino 2003). As a result, we are likely to underestimate inline image by assuming that Ethiopian wolves’ diet has no plasticity. It is therefore possible that by changing their diets in response to diminishing rodent densities (by either eating less prey at low prey density or by switching prey), Ethiopian wolves may be more resilient to an increase in livestock density. We would expect any dietary plasticity to produce a trophic response for wolves intermediate between type 1 and type 2, the two extreme scenarios considered here. However, such diet plasticity is expected to be limited considering the wolves’ high degree of specialization on rodents.

The absence of long-term baseline monitoring of rodent densities and vegetation condition precludes any assessment of changes in the state of this ecosystem since the advent of the high grazing pressures recorded on parts of the Ethiopian wolf range in BMNP. The ability of these populations to recover from disease outbreak is limited by the formation of breeding units, and subsequent recruitment of young to the population (Marino, Sillero-Zubiri & Macdonald 2006), itself largely determined by prey abundance (Tallents 2007). Historically, the Web valley has supported the highest wolf densities in the BMNP and wolves there have until recently recovered from periodic rabies outbreaks suggesting that past conditions have supported positive wolf population growth rates. However, the models suggest that the long-term persistence of wolves there (and in Morebawa) may not be compatible with current grazing pressures.

This projection is evidently at odds with the continuing presence of wolves in the Web valley and Morebawa, and there are two possible explanations for this. The first is that the models structure or parameters are simply wrong. Given the absence of more detailed information, analysing a simple model is a useful prerequisite to the inclusion of greater complexity and it remains possible that a more detailed treatment of primary production, seasonality, space and the taxonomic grouping of plant and rodent species might lead to different conclusions. The second is that the system is in transition and moving towards the state that the models predict given the heavy grazing pressure. Long-term transients are an anticipated property of some types of ecological system (Hastings 2004), but all the models analysed here predict a time to extinction of <5 years, with models that assume type 1 functional responses taking longer than those that assume type 2. We are, however, cautious in placing too much confidence in these transients, which may very well be sensitive to assumptions about seasonality and the lack of an explicitly spatial dimension.

Management implications

We have described a quantitative approach for assessing the resilience of afroalpine food webs to disturbance by livestock grazing. Although this approach is applied to a particular food chain in the BMNP, the model could be parameterized for other afroalpine systems that sustain Ethiopian wolf populations or for which empirical data on standing crops and biomass flows between the different trophic levels are established. The models could also be modified to look at the effects of other types of disturbance, such as fire for example (Wesche 2002; Clausnitzer 2003; Layme, Lima & Magnusson 2004), which may reduce the afroalpine vegetation standing crop and productivity and impact on the rodent population.

Our findings have two implications. First, the monitoring of species at the top of food chains may not always be informative of the ecological changes happening at lower trophic levels. Our trophic models show that if consumer species exhibit (extreme) type 2 functional responses, their populations may be maintained close to unperturbed equilibrium levels for a long time even when vegetation biomass has been greatly reduced as a result of grazing disturbance by livestock. Such dynamics make it difficult to predict population crashes and/or local extinction of Ethiopian wolves. This suggests that the emphasis of current monitoring on species-specific population indices might be extended to include metrics which are likely to be more informative of the dynamics of key ecosystem functions. For example, while the monitoring of Ethiopian wolves remains important for early detection of disease outbreaks, monitoring afroalpine vegetation productivity might allow managers to foresee dramatic shifts in the stability of afroalpine trophic chains. Finally, we have shown that stability analysis can be used to inform policies on the mitigation of human impacts on the environment by predicting which areas have been pushed into an unstable configuration and those that will be particularly sensitive to future increases in the level of disturbance.


We thank the Ethiopian Wildlife Conservation Department, the Oromia Regional Government and the Bale Mountains National Park for permission to undertake this research and conservation work. We are grateful to the Ethiopian Wolf Conservation Project and Frankfurt Zoological Society for their logistical support and dedication in the field. Funding was provided by the Royal Geographical Society with a British Airways Travel Bursary; by the Wildlife Conservation Society under the Research Fellowship programme and Frankfurt Zoological Society.