Does restoring apex predators to food webs restore ecosystems? Reply

Beschta and colleagues offer a lengthy, wide-ranging critique of Hobbs et al. (2024) in their Comment (Beschta et al., 2025). We focus our Reply to Beschta et al. on the overarching issues they raise, that the scope of inference of our paper is not supported by our research and that we failed to adequately consider extensive evidence of a trophic cascade initiated by the restoration of wolves to the food web of the northern range of Yellowstone National Park. It is useful to summarize the line of inference developed in the Hobbs et al. monograph. The paper contributes to ecological theory by addressing the core question: “Can the simplification of food webs caused by the removal of apex predators create conditions in the ecosystem that are resilient to the effects of predator restoration?” using the small stream network of Yellowstone's northern range as a model system. The return of a guild of large carnivores to the northern range of Yellowstone after their protracted absence offered an unusually valuable opportunity to address this question. The state of the ecosystem along the network of small streams changed dramatically during the twentieth century as we describe in the section on History. These changes appeared to be initiated by the loss of apex predators from the system. In particular, tall willows were mostly eliminated from stream sides and engineering of streams by beaver ceased. Established ecological theory on hysteresis and the empirical results of Wolf et al. (2007) (summarized in Figures 1, 2, and 3 of Hobbs et al.) motivated two competing hypotheses on the stability of the small stream ecosystem in response to the disturbance created by restoration of top predators to the food web (Figure 5B,C and the section Experimental Design). The first hypothesis specified that the system failed to exhibit hysteresis, that the restoration of apex predators allowed a rapid return of the conditions that prevailed on the small stream network before the removal of apex predators, especially the presence of tall willows (Salix spp.) and engineering by beaver (Castor canadensis). The second, alternative hypothesis was that the removal of apex predators created hysteresis in the system by interrupting the symbiosis of willows and beaver. These hypotheses were tested in a replicated, manipulative experiment and analysis of a set of sites drawn from areas of the small stream network suitable for colonization by beaver. Analysis of two decades of data from this experiment allowed unambiguous rejection of the first hypothesis, leading to the conclusion that the conditions that emerged while predators were absent from streams capable of being engineered by beaver were not quickly reversed by predator restoration. The conclusions of Hobbs et al., 2024 extended from development of hypotheses motivated by established ecological theory, from our experimental design, and from our rigorous statistical analysis of the experimental data. Beschta et al. might disagree with these conclusions based on their own research, but there is nothing in their Comment that shows errors in the line of inference of Hobbs et al. (2024). We emphasize that the strength of our inferential approach was that we evaluated evidence for alternative ideas about how an ecosystem, the small stream network of the northern range, responded to a perturbation, the restoration of apex predators. The consideration of alternative explanations for states and processes is the heart of reliable science, a truth that we will soon revisit. “Our two decades of research revealed that the restoration of apex predators to a food web after a long absence failed to reverse the effects of their loss from the food web.” “Our two decades of research revealed that the restoration of apex predators to a food web after a long absence failed to reverse the effects of their loss from the food web. This absence of reversal could be attributed to effects of slowly changing ecosystem conditions (i.e., “parameters” in the language of Beisner et al. (2003)) that emerged while predators were absent from the food web. The elk–grassland state along the small stream network of the northern range in Yellowstone National Park bore the hallmark of an alternative stable state (Folke et al., 2004; Petraitis, 2013; Scheffer et al., 2001): a change in the food web caused the ecosystem to reconfigure to a new state characterized by dramatic shifts in plant communities, an altered disturbance regime, and long-lasting changes in the physical environment.” The remainder of the paragraph makes clear that our conclusions about “failure” were limited to the alternative hypotheses we developed from our Conceptual Framework, specifically that restoration of apex predators has not caused the elk-grassland state to reverse to the beaver-willow state on the network of small streams, ideas that are justified in detail in our Figure 2 and associated discussion. Beschta et al. assert that the scope of our inference depends on an arbitrary definition of restoration. We find this criticism without merit. We justify our definition of restoration in our Figure 2 and the text associated with it. There is nothing arbitrary about this justification. Beschta et al. also contend that our inference should be limited to the 30% of the reaches of the small stream network that we studied and does not inform sites that historically were not occupied by beaver. This contention is true in a strict statistical sense, but such a narrow limitation on our inference neglects the well-established understanding of the disproportionate effects of engineering by beaver on ecosystem states and processes (Larsen et al., 2021; Naiman & Décamps, 1997). Effects of localized disturbance by beaver extend across entire watersheds. A degraded ecosystem cannot be considered restored in the absence of the restoration of its disturbance regime. The arguments of Beschta et al. about our excessively broad inference extend from statements about our study design that are inaccurate. Beschta et al. write that control sites in our experiment were chosen to represent “dry conditions”, which distorts our criteria for description of selection of sites (see Experiment Design). Moreover, these criteria were applied equally to treatments as well as controls, not to the controls alone, as Beschta et al. write. They also argue that our experiment “…excluded the most resilient, diverse, and productive streamside sites where willows are supported by groundwater or high-water tables, missing key elements of processes important for restoring the beaver-willow state.” We chose to work in sites without high-water tables created by groundwater inflows to the streams so that we could test the importance of water availability on willow growth and the role of the disturbance regime created by beaver replicated by our simulated beaver structures. In addition, many groundwater-fed sites are peat accumulating fens and are not riparian ecosystems. Their statement is narrowly true, but it overlooks the broad results of our analysis (Table 2, Figures 14 and 15) where we show that low variability in availability of groundwater accelerates willow growth. A novel contribution of our work was to show that the strength of hysteresis depended on landscape variability in availability of groundwater able to support willow growth (Figure 21 and associated discussion). It follows that we used our design to make precisely the point that Besctha et al. said we were unable to make. We show (Figure 21) that the beaver-willow state is rapidly restored in areas with highly available groundwater. Beschta et al. do not speak to the overarching goal of our paper. We sought to advance ecological theory by considering when we would expect the restoration of apex predators to rapidly reverse the effects of their loss, a phenomenon we called reciprocal effects. We developed these ideas thoroughly in Figure 1 and its associated discussion, predicting that simple food chains would be expected to show reciprocal effects, while complex food webs would not. We used Yellowstone as a model system to advance these ideas, showing that our results are consistent with the prediction that degradation of ecosystems caused by the absence of predators may not be quickly reversed by predator restoration because changes in ecosystem states and processes that occurred in the absence of predators, notably changes in the disturbance regime, create hysteresis. All investigators who seek to use empirical studies to advance theory must describe how the specific system they studied provides general insight. Individual readers can judge if the general insight we provided to advance theory exceeds the support provided by the data and the analysis. Beschta et al. raise many concerns that we failed to acknowledge extensive evidence that the reintroduction of wolves initiated a trajectory of recovery of woody deciduous plants via a trophic cascade. In our defense, we did review all of the studies of willow growth along small streams after the reintroduction of wolves, including those that advocated a trophic cascade. We stated clearly that these studies reached conclusions that differed from ours. We offered reasons for these differences (see the section in our paper “Comparison with other studies of willows in the northern range”). We did not review studies of other plant communities and vertebrate consumers that Beschta et al. argue we should have included because we judged those studies to be tangential to core ideas in our paper. However, a more important motivation for omitting those studies was that we concluded they offered unreliable inference. Our conclusion was based on two lines of reasoning: the importance of experiments to causal inference and the requirement to consider alternative explanations for trends in plant growth following the reintroduction of wolves. Manipulative experiments investigating trophic cascades in terrestrial, vertebrate food webs have revealed that indirect benefits of predator on plants were minimal or absent (Maron & Pearson, 2011). The fundamental importance of experimentation to reliable inference on growth of woody deciduous plants following the restoration of predators to the northern range is illustrated by pretending for the moment that the data from our control and observational sites were the only observations we collected. What would we see? There is a clear trend of increase in willow height at most (not all) of these sites during the years following predator restoration. We showed that willow growth rate was negatively correlated with browsing intensity (Figure 15). It would be tempting to conclude from these temporal trends and correlations that wolves and perhaps other predators initiated the recovery of willows by causing reduced browsing. However, when we add the data from the designed, experimental manipulations, we find a much more complex picture. We find that removal of the initiating stressor, the absence of apex predators, failed to restore the conditions that prevailed when the food web was complete. Restoration of predators did not allow willows to grow nearly as rapidly as they would be expected to grow in a restored ecosystem, thereby demonstrating hysteresis. We acknowledged (page 22 paragraph 2 of Hobbs et al.) that willows may ultimately reach the heights expected in a restored ecosystem given sufficient time, but we also offer several credible, alternative futures for willow communities that do not include the conditions that prevailed before wolves were reintroduced (page 22 and 23 Hobbs et al.). All of the studies that Beschta et al. argue we wrongly omitted from our manuscript share a common weakness. The evidence showing a trophic cascade in these studies is observational and correlative, suffering from the same problems of inference that we would have experienced in the absence of our experimental manipulations. These weaknesses are on full display in the analysis of willow crown volume described in Beschta et al. using our data (their fig. 1). Data on crown volumes of our dam and fence treatments were deliberately omitted from that analysis. We suspect that including dam and fence treatments in their analysis of the temporal trend in crown volume would have shown the same differences as those we show for height, given the algebra in the model they used to predict volume from height. Crown volume was used as a surrogate for estimates of accumulated biomass (Beschta et al. fig. 1), but Beschta et al. fail to point out that Marshall et al. (2013) measured accumulated biomass and found results consistent with those in our monograph. MacNulty et al. (2025) showed fundamental flaws in the analysis of willow crown volume by Beschta et al., including a tautological volume model, violations of key modeling assumptions, comparisons across unmatched plots, and selectively framed photographic evidence (also see Bilyeu et al. (2008) fig. 1). The papers Beschta et al. argue we should have cited lack a null model that is compared with an effects model. If wolves or other predators have caused accelerated growth of willows and other woody deciduous plants, what is the “no wolf” growth rate? Zero growth? Half the observed growth rate? Observational methods alone provide no way of knowing the rate at which plants would have grown if apex predators were absent and therefore provide no way to assess the magnitude of the predator effect. Assuming that any observed growth after wolf reintroduction is attributable to the indirect effects of wolves requires the assumption that willows could not have grown tall in the absence of wolves. Research on willow heights conducted before the reintroduction of wolves casts doubt on this assumption (Singer et al., 1994), as we pointed out in our paper (page 21 paragraph 2). The conclusions of the papers that Beschta et al. contend we wrongly omitted from our monograph failed to consider these alternative explanations alongside the single explanation they favored, a wolf-induced trophic cascade (also see Marris (2014)). Instead, this body of work inferred causation from precedence to conclude that a trophic cascade is responsible for observations of patchy plant growth in the years after wolf reintroduction. We stand by our decision to omit the conclusions of these papers from our monograph. The seminal hypothesis that reintroduced wolves initiated recovery of the plant communities of Yellowstone's northern range by causing a trophic cascade mediated by predation risk (Ripple & Beschta, 2004) captured the imagination of scientists and the public alike. It was a beautiful idea. Science confronts ideas with evidence obtained by proven methods, notably manipulative experiments, statistically robust and appropriately designed sampling, sanctioned analysis, and evaluation of competing explanations for states and processes. More often than not, science reveals that ecosystem processes are more complex than initially believed. Our monograph adds to the large body of evidence revealing that the events that played out after the reintroduction of wolves to Yellowstone were far more complex than was anticipated by the trophic cascade hypothesis (Bilyeu et al., 2008; Brice et al., 2022; Brice, Larsen, et al., 2025a; Creel & Christianson, 2009; Eisenberg et al., 2013; Kauffman et al., 2010; Kohl et al., 2018; MacNulty et al., 2024; MacNulty et al., 2025; Marshall et al., 2013; Marshall et al., 2014; Peterson et al., 2014; Ruth, 2019; Stahler & MacNulty, 2020; Vucetich et al., 2005; Winnie, 2012). Embracing the complexity of the interactions that followed the reintroduction of wolves to Yellowstone does not mean wolves had no influence. Rather, any effects of wolves on the trajectory of the ecosystem occurred in concert with effects of human actions, predation by other large carnivores, weather, and changes in the disturbance regime and physical environment that occurred when wolves were absent. The weight of evidence shows that wolves are one component of a suite of biotic and abiotic controls on recovery of woody deciduous plants on the northern range. The authors declare no conflicts of interest.