Extinction cascades, also known as secondary extinctions or coextinctions, are extinctions that occur as a result of another, “primary”, extinction. They’ve been a popular topic for many years in the ecological “network” literature. For instance, if you have a food web (a “network” showing who eats whom), or a plant-pollinator web (a “network” showing who pollinates whom), you can remove species from the network and see whether extinction cascades result, under some assumptions about how the ecology of the system works.
I used to be up on the modeling literature on extinction cascades, but gradually stopped following it closely as my interests shifted away from food webby topics. But recently I read the new review of the theoretical and empirical literature on extinction cascades by Colwell et al. (2012). I was curious to come back to this topic after some time “away”, to see how things had changed. Extinction cascades are both fundamentally interesting, and potentially of considerable applied importance, so I wanted to see how the literature had developed. Reading Colwell et al.’s review was useful, but left me with some questions. That’s not a criticism of the review, I just have questions because I think about what I read. So I decided to turn those questions into a post, to see if readers who know more than I do can address them. I know lots of folks read this blog to learn something from me, but sometimes (as with my post on SEMs) I use it as a way to learn from readers. And even if you can’t help me out with my questions, hopefully what follows will be an interesting example of someone “thinking out loud” about the current literature.
1. Reading Colwell et al., I was struck by just how limited the empirical evidence on extinction cascades is. According to Colwell et al., there aren’t many well-established empirical examples of extinction cascades at all, and the few that we have mostly involve highly-specialized ecoparasites going extinct when their one and only host goes extinct. And what indirect evidence there is seems like fairly weak sauce to me. For instance, introduced species often lack some of the parasites, predators, and pollinators from their native ranges–a pattern that could well arise from the absent species just so happening not to have been introduced along with the focal species. Extinction of obligate specialist parasites when their only host goes extinct is the minimum possible extinction cascade–a host goes extinct, and so do its obligate specialist parasites, with no further knock-on effects. While that may be an important phenomenon, it seems rather different than the sort of thing often considered in the modeling work with which I’m familiar. The modeling literature that I used to read focused on more exotic possibilities like numerous primary extinctions reducing a previously-generalized food web to a bunch of linked specialists, which then collapses entirely after one more extinction, like a Jenga tower from which one too many blocks have been removed. Is this just me? Should I be surprised that the empirical evidence on extinction cascades is limited? If we don’t have empirical examples of the more exotic sorts of extinction cascades, why is that? And if those more exotic sorts of extinction cascades truly don’t happen, does that mean a lot of our modeling work on extinction cascades is rather detached from empirical reality?
2. Then again, maybe there are lots of good examples of extinction cascades–just not the sort of extinction cascades contemplated by the existing literature. Bob Paine (1966) found perhaps the classic example of an extinction cascade, which he termed keystone predation. He removed Pisaster seastars from a patch of rocky intertidal, and the patch was eventually taken over by blue mussels that, in the absence of selective predation on them by seastars, were able to outcompete everything else. But I have the impression that the extinction cascade literature mostly (not entirely, but mostly) ignores this sort of possibility. The theoretical and empirical literature reviewed by Colwell et al. seems very focused on secondary extinctions that occur when some species loses all of its prey, or all of its hosts, or all of the mutualists on which it depends. The possibility of secondary extinctions due to dynamical processes like competition, apparent competition, etc., appears not to be much considered in the extinction cascade literature. Indeed, examples like Paine (1966) apparently aren’t even regarded as extinction cascades at all: in reviewing the empirical literature on extinction cascades, Colwell et al. (2012) don’t cite Paine (1966), or any other removal experiment from the community ecology literature on diversity and coexistence. That strikes me as quite a narrow focus–a focus on only one particular sort of extinction cascade–but I suspect I’m missing something. I guess there must be some motivation for this narrow focus that I’m not getting–can anyone help me out?
3. Back when I was following the extinction cascades literature, the models I knew about were “network” models. (Colwell et al. also discuss what they call “statistical” models of extinction cascades, but I won’t get into those) Back when I was up on the literature, network models of extinction cascades mostly were based on simple but unrealistic assumptions about how primary and secondary extinctions work. I have the sense from the Colwell et al. review that some of those assumptions are getting relaxed, but only some of them:
- It used to be that network models of extinction cascades all assumed that “network rewiring” is impossible. That is, they assumed that the observed predator-prey or plant-pollinator interactions are the only possible ones. While that’s a natural starting point, it is of course unrealistic in most cases (and not just because observed networks invariably are undersampled and so omit rare interactions, although they are and do). Further, it’s especially unrealistic for species that have recently lost some of their prey, or plants, or pollinators. That is, it’s especially unrealistic in the very situations that we’re most interested in if we’re trying to model secondary extinctions. Here’s a beautiful illustration of the lengths a predator will go to, and the changes it will make to its usual diet, to avoid starving when its usual prey are no longer available. So I was glad to see that more recent network models of extinction cascades are getting away from this assumption and allowing network rewiring (i.e. adaptive changes in who interacts with whom; e.g., see Valdovino et al. 2010 for a review of adaptive foraging models in complex food webs). Although even there, it looks like many models still assume that the observed interactions are the only possible ones? If so, that still seems like an unrealistic assumption, and it seems like it would be interesting to try to relax it. The work of my friends Owen Petchey, Andy Beckerman, and Phil Warren is a nice example of work that avoids this unrealistic assumption (see Petchey et al. 2008 PNAS), though it’s still early days and it’s unclear to me how far the Petchey et al. approach can be pushed. Is the modeling literature indeed moving in the direction of “network rewiring” models? Particularly ones that allow for the possibility of species responding to extinctions by starting to eat, or pollinate, species they’ve never previously been observed to eat or pollinate?
- Another common assumption of the older modeling literature is that primary extinctions are assumed to occur one by one, and to occur via species simply “vanishing” for unspecified reasons. Am I right in my impression that that’s still a common assumption these days? That approach to modeling primary extinctions is undoubtedly the simplest starting point. It has a venerable history in food web modeling, going back at least as far as Stuart Pimm’s classic modeling studies of “species deletion stability” in the 1980s. (Aside: “species deletion stability” is a small but relevant body of modeling work that doesn’t seem to be much discussed in the recent extinction cascade literature; it’s not cited by Colwell et al…) Assuming that primary extinctions occur via species just “vanishing” one by one (either at random, or in some pre-specified order) is probably ok if your sole focus is trying to understand the conditions under which extinction cascades will occur. But it’s surely an unrealistic assumption. Extinction is a dynamical process–species don’t just vanish (well, unless they’re experimentally removed). They first decline in some fashion, and that decline itself has dynamical consequences for the other species in the community. Declines can even lead to secondary extinctions without the declining species themselves actually going extinct. It’s not at all clear to me that the dynamical response of a community to a dynamical extinction will mimic its dynamical response to species simply vanishing. Has anyone looked at that? Further, when I think about systems in which there are lots of primary extinctions, I can’t help but think about the major causes of those extinctions–things like habitat loss and modification. Causes that affect all species simultaneously rather than one at a time, and that seem likely to obscure or even overwhelm extinction cascades arising solely as a response to primary extinction. I don’t question the value of extinction cascade modeling as a starting point, or as a purely theoretical exercise (theory does have a life of its own, independent of data). But I do wonder if existing work isn’t missing some really interesting and important possibilities in not considering realistic causes and dynamics of primary extinctions.
Again, I emphasize that this post is about me asking questions. I’m not critical, I’m curious. Looking forward to comments.
Thanks for highlighting the article, Jeremy. I’m a bit disappointed with some of the literature they’ve missed out (aside from my bruised ego at not being cited at all – but if Stuart Pimm doesn’t get a look in, what hope do I have?), they missed the recent and very relevant Zavaleta et al (2009) review Ecosystem Responses to Community Disassembly, which covers a bunch of the empirical evidence that Colwell et al suggest is lacking.
I agree that there isn’t enough high quality empirical/experimental evidence about this stuff out there and that it’s a hugely challenging topic to deal with in the real world, but that’s quite an important article to miss from the reference list. I’ll reserve judgement about the rest of the Colwell et al article until I’ve read it carefully.
You could argue that all harvesting models and experiments start to do this to some degree. At least this one sort of makes a start.
In fairness to Colwell et al., the Zaveleta et al. review is focused on the effects of extinctions (cascading or not) on ecosystem function. So I can see why Colwell et al. wouldn’t cite them (though to the extent that Zavaleta et al. cite studies of extinction cascades that Colwell et al. don’t, that does seem odd.)
Good point re: harvesting models as models of dynamical responses to species decline.
I’m not a big food web guy, but I do think quite about species networks and ecosystem functioning. I was immediately reminded, in reading the first few sentences of your post, of Bob Paine’s work on keystone species. I’m glad that you brought it up and this seems like a great example of an extinction cascade.
I was struck second thinking about about the difficulty in showing an extinction at all. In the key stone species approach is relatively easy to demonstrate a local reduction in density or even local extirpation. But a complete extinction would be difficult indeed to show empirically! A corollary example is the work by Sax asserting that no invasive species has been empirically shown to cause a species extinction. Negative effects of invasives can be demonstrated, but how do you show that the very last individual of a species has left this earth? Sampling issues alone would make it impossible. Confounding issues of simultaneous changes caused by humans (assuming humans are the agent of extinction here) such as habitat fragmentation that could also be negatively affecting species makes it difficult to show direct causality.
Other than a model or mesocosm world, it seems like the best hope of showing an extinction cascade is one where the agent of extinction were a disease that only eliminated one particular species. In this hypothetical scenario, if a disease wiped only one taxa out and you could then look at the changes to the system following that taxa’s extinction (local extinction at least…). Chytrid fungus driven extinction of frog species offers one potential study system that has these characteristics, although that removes multiple species rather than just one….
You’re certainly right that it’s very difficult to conclusively prove when something has gone extinct. But I’m not sure that explains the apparent rarity of extinction cascades in the Colwell et al. review. I mean, if you see extinction cascades as rare because you only want to consider cases where we’re sure species have gone extinct, then wouldn’t you just see extinctions as rare? Put another way, if you’re willing to accept strong circumstantial or indirect evidence that lots of primary extinctions are occurring, thereby causing you to worry about the secondary extinctions that those primary extinctions might lead to, why not accept the same sort of evidence when it comes to secondary extinctions?
And while it may be difficult to show that the last individual of an entire species has gone extinct, it’s much easier to show that local populations have been extirpated. And the possibility that local extirpations will result in local extinction cascades affecting local populations of other species seems like a pretty important possibility to consider. This is of course the sort of “local” extinction cascade demonstrated by field experiments like that of Paine (1966). Maybe Colwell et al. want to restrict attention only to cases where entire species go extinct; I’m not sure. But if so, why? Seems to me that “local” extinction cascades could potentially be important. If nothing else, studying them should tell you something about how “global” extinction cascades work (since “global” extinction is simply what happens when the last local population is extirpated).
I absolutely don’t mean my comment here as a criticism of your points, which are very well-taken. My comments are really just to re-emphasize my confusion about the literature Colwell et al. choose to review, and the literature they chose not to include. It’s not just that their review seems narrower than a typical AREES review, the choices they made still confuse me even assuming that they wanted a narrow focus. I’m trying to decide whether my confusion reflects a problem with me, a problem with their paper, or a problem with the broader literature on this topic.
Yes. And by yes I mean I agree with your points and haven’t read Colwell’s paper, so I can’t contribute to speculation about what their intentions were when formulating the review. It does seem, based upon your description, and my now, as I write this, skimming Colwell’s paper that they intended to focus on examples of global extinction rather than the comparatively easy challenge to demonstrating local extinction cascades and scaling up the inference….
In thinking about ecological networks and how we often study them it may not be surprising that we rarely see extinction cascades in the empirical literature. Usually in the theory papers we see networks treated as primarily static objects, where much of the data is averaged across numerous years. That does not really accurately represent reality. If you read the (admittedly few) papers that have empirically examined the temporal dynamics of interaction networks (e.g. Olesen et al. 2008) they are often characterized by high species turnover. If that is the case, then it would not be unexpected that if one species were to go extinct it would be replaced relatively quickly by something else. In other words we cannot think of these networks as closed systems but rather as a subset of a much larger regional (maybe even global) web with many species coming and going depending on their movement/speciation/dynamics/etc.
Good point Jon. I actually am familiar with some of the empirical literature on temporal dynamics of networks, as I have a grad student working on that topic in plant-pollinator networks. You’re right that we need to think about larger scale, metacommunity-type systems in order to think about this issue properly. There is theoretical literature on this, I believe, but it’s not a big literature by any means. You could well be right that one reason why even “local” extinction cascades are rare is because, when at least some members of the network are mobile, they can just move elsewhere rather than go locally extinct.
Interesting post and a topic I’ve been think about a lot recently in relation to plant-pollinator networks and I think you’re right, Jon. The static nature of the (rather few) published simulations, in which re-wiring is not allowed, prompted us to perform an experiment in which we (i) characterised the network of a grassland p-p assemblage; (ii) removed the most generalised plant within the network (which involved hand pulling over 14,000 flower heads in a single day!); (iii) re-assessed the network after a couple of days; (iv) assessed the network after the most generalised plant had re-grown.
The result? Removing the core resource made no difference to the abundance and diversity of pollinators in the network. All that happened was that the abundance of pollinators visiting the previously second most generalised species increased, though that effect did not propagate down through to the other plants. And once the removed species re-grew, the pollinators moved back on its flower heads and it re-established itself as the most generalist species.
We also monitored a nearby grassland as a control and found that there was no increase in pollinators on that site. We had expected the pollinators to vote with their wings, as per Jeremy’s comment that “when at least some members of the network are mobile, they can just move elsewhere rather than go locally extinct”. But there was no evidence that this had happened.
I don’t think that this tells us much about extinction cascades, but it does show how resilient plant-pollinator networks are (over the short term) because of the behavioural flexibility of the pollinators.
I have a manuscript detailing the study that is more or less ready to be submitted.
Sounds like a nice experiment Jeff. Certainly jives with my own personal view that, unless you’re modeling extinction cascades as a purely theoretical exercise (in which case, you need to recognize that that’s what you’re doing and not pretend that your model is empirically-grounded), there’s really no point in looking at a model that’s non-dynamical, and that doesn’t allow network rewiring (including both quantitative shifts in pollination or feeding rates, like you describe, and qualitative shifts involving pollinating or eating things you’ve never eaten or pollinated before).
The issue of when a consumer or pollinator will shift to eating or pollinating something else at the same site, vs. moving elsewhere in order to keep eating or pollinating the same things, is an interesting one. Probably a fair number of people have studied it (both theoretically and empirically), but I just haven’t seen those studies or have forgotten them. I recall that the wife of my undergrad advisor did her PhD on moose foraging behavior on Isle Royale. Moose there spend the spring and early summer eating lichen growing on trees on Isle Royale. But by about early-to-mid July, they exhaust this forage. Instead of switching to (presumably) less-desirable food, they start swimming around to all the little islets surrounding Isle Royale and eating the lichens there. Now, moose are pretty good swimmers as terrestrial tetrapods go, but still, swimming among islets seems like a pretty energetically-expensive choice to me! But I assume that the moose must more or less know what they’re doing.
Regarding the moose, I wonder if boredom played a role? Has anyone ever considered boredom as a factor promoting mammal (or bird) movements? We know from captive studies that animals constrained to small areas display behaviours consistent with what we would call being bored. Just a wild thought for a Friday morning 🙂
If moose on Isle Royale are moving around because of boredom, I would guess they’re seeking boredom rather than trying to relieve it. There are wolves on the main island, but not on the islets. Perhaps by leaving the main island, moose are seeking a quiet, boring life! 🙂
Ha ha, that’s a good point.
Theoretical studies of secondary extinctions generally have used either topological or dynamic criteria. A topological criterion is that you (supposing you are a population) go extinct if all your prey are extinct (e.g., Dunne et al 2002 Ecology Letters). A dynamic criterion is that you go extinct if your abundance goes below a certain value. Topological criteria just look at the food web topology, dynamic require some kind of dynamic model. Bo Ebenman et al have done a lot of work on secondary extinctions using dynamic models, and showed that topological criteria can badly underestimate the number of secondary extinction.
By the way, take a look at this for an illustration of the effect of assuming finite (stochastic) or infinite (deterministic) populations on secondary extinctions.
Two papers about the effect of rewiring on the prevalence of secondary extinctions:
Thierry et al (2011) Basic and Applied Ecology
This one uses optimal foraging theory (you cite the PNAS paper) to rewire the food web after an extinction. The criteria for a secondary extinction is based on the reduction in energy intake rate. If even with rewiring, your energy intake rate is < x% of the original, you go extinct. Rewiring reduces secondary extinctions and changing x acts as you would expect.
Staniczenko, P. P. A., Lewis, O. T., Jones, N. S. & Reed-Tsochas, F. (2010). Structural dynamics and robustness of food webs. Ecology Letters, 13, 891–899.
This uses a rather phenomenological rule to rewire, with quite predictable effects (i.e., decreases in secondary extinctions).
I can't really comment on the review you post about, as I haven't read it (yet).
Cheers for this. I of course know some of this work, Bo Ebenmann’s stuff in particular (though even that assumes non-dynamical primary extinctions).
Pingback: On the tone and content of this blog (feedback encouraged) | Dynamic Ecology