The metacommunity concept: a framework for multi-scale community ecology

Leibold et al. (2004) Ecology Letters 7: 601-612. DOI:10.1111/j.1461-0248.2004.00608.x. The metacommunity concept: a framework for multi-scale community ecology


Naughtily, this is a diagram of (roughly) the same concepts as discussed in this paper but from Logue et al. (2011). NM: Neutral Model, PD: Patch Dynamics, ME: Mass Effect, and SS: Species-Sorting.

Will Pearse

Will Pearse

What surprised me most about this paper was how much of it I feel I have absorbed, and yet I can’t consciously recall reading it. It’s a classic in the field, and I think either influenced or consolidated a lot of what people thought about metacommunity structure. It’s a great paper, and if you can’t recall reading it I suggest you go ahead and do so.

I don’t want to dig up old ground, but I was pleased to read the authors making explicit claims about how different processes would be picked up depending on the evolutionary history of the system. It’s great to see an attempt at integrating fields (when was the last time you heard someone call Neutral Theory a metacommunity model?) that doesn’t just stop at the line of ecology. Last time we discussed whether species truly neutrally dispersed, and how dispersal traits can interact with traits that we consider in a classic ‘here’s my quadrat what’s growing in it’ ecology. Metacommunity dynamics open up a whole range of additional processes and evolutionary interactions that can be simulated and estimated using empirical data – although whether we actually do that is a different question.

The authors claiming not to have covered spatially explicit models got me thinking. When we say ‘spatially explicit’, we typically mean ‘each individual has an x,y(,z) co-ordinate which we model’, and these models can be very difficult to fit. I think the authors are right that we don’t always have to use such models to capture interesting dynamics – three levels of hierarchical spatial nesting are often enough for me! However, if we were to fit a spatially explicit model over a large enough area, with different habitat types and dispersal across the entire space (perhaps separating between long-distance and short-distance dispersal), we should essentially be able to replicate metacommunity dynamics. I don’t think I’m alone in saying that, while there is a metacommunity, there’s no real such thing a community – it’s just what individuals happen to be in the unit that we’ve defined at that point in time to be useful for us to study something of interest (here’s some Vellend). It’s communities all the way down, each capturing a different scale of interactions or species, and perhaps we would have a better chance of capturing such dynamics if we examined whether we can get meta-community-like behaviour emerging natural from spatially explicit models. In passing, for every person who emails/comments screaming about how communities are real, I will donate $1 to the ‘I made a sweeping statement sorry everyone’ fund.

Lynsey McInnes

Lynsey McInnes

Contrary to Will, I found this paper tough-going. Not because it was bad, uninteresting or poorly written, probably just because it was extremely dense. And my mind constantly kept wandering and wondering – was this really published 10 years ago? How have we moved on from here?

I’ve always had a soft spot for meta- type models while never knowing many of the details. But from my ill-informed sideline position, I don’t really feel like we have moved on much from this landmark paper. Have we? Correct me if I’m wrong.

So, that nagging feeling led me to wonder why we might not have moved on much? Is it a data availability thing? A model availability thing? A every collection of ‘communities’ is different thing? Or what? Ja, ja, it’s probably just a combination of all three and more.

So, where would I like to see things go? Well, unlike me, I think we need to spend more time working out what makes a metacommunity ‘real’ before we can really tackle how it fluctuates through space and time. Maybe a good place to focus would be working out what populations within a ‘community’ interact, how stable or transient these interactions are and then add in links to neighbouring communities and quantify how strongly connected they are. I say – use genetics! Use the genome. Let the populations tell you how they are related to each other. Fit admixture models. Fit migration models. See how congruent models are among populations. Sure, this perspective is limited to a distinct time band, it won’t work for really transient metacommunities, but it will work for established ones and could help identify which populations are stable within (meta)communities and which fluctuate in importance and could lead to more informed models for faster-turnover metacommunities. If we use genetics to let populations speak for themselves, we also won’t go wrong if we add another layer of complexity and incorporate trait variation. We might be considering six communities, each with an overlapping set of species, but spatially-distributed populations of the same species will not have the same trait complement. Recognise this! Quantify it! Find out how it happens and why it matters!

No doubt these models are already been fitted, but how much crosstalk is there between pure ecologists, metacommunity ecologists and population biologists on the one hand and geneticists on the other hand. Let’s integrate!

My big dream is for us to one day understand how diversity gets organised from the scale of individual interactions through community dynamics to shifting ranges and ultimately species’ turnover. We will not get there without more communication from the people best placed to understand the processes occurring at each scale. The metacommunity concept is a great place to start as it links individuals, populations, trophic interactions and communities. We just need to use the best data to make inferences about all of these.
*Apologies for the rushed, overly exclamation-marked rant… Metacommunities are a great concept, let’s see how far we can push them. (And apologies if all this integration has happened and just passed me by…).


About will.pearse
Ecology / evolutionary biologist

4 Responses to The metacommunity concept: a framework for multi-scale community ecology

  1. Jeremy Fox says:

    No, community ecology hasn’t moved on from this paper, at least not much. In part because this paper quickly became the basis of a popular empirical research approach: take observational data on which species are found at which sites (and perhaps at which abundances), along with data on environmental variables, run them through a variance partitioning analysis, and from that try to infer which of the four classes of metacommunity model identified by Leibold et al. describes the system. Indeed, some authors even write as if the variance partitioning approach *describes* or *defines* what type of metacommunity you have (e.g., some authors define a “species sorting” metacommunity as one in which environmental variables explain lots of variation in species composition while spatial structure explains little). Soininen (in press at Ecology) reviews this variance partitioning work.

    Unfortunately, this approach is problematic, for various reasons. For instance, Leibold et al. never claimed to, and didn’t, identify all possible kinds of metacommunities. They were just summarizing and organizing the theoretical literature that existed at the time. There are actually other sorts of metacommunity dynamics that ecologists have never modeled (though population geneticists have): The problem here goes beyond nature possibly being intermediate between the various sorts of idealized situations theoreticians have modeled (that’s a widely-recognized issue).

    Another problem with the variance partitioning approach is that the few people who’ve tried applying it to simulated data generated by known processes have found that it often fails badly at recovering those processes. That is, our intuitions about which sorts of metacommunities should be characterized by which sorts of variance partitioning results seem to be off base (Gilbert and Bennett 2010, Smith and Lundholm 2010). Arguably, you see a signal of this in the empirical results reviewed by Soininen. Those results are basically a shotgun blast–for instance, you don’t see any difference on average between metacommunities of more- and less-mobile species, as you’d expect intuitively. Which might be a sign that our intuitions aren’t well-grounded.

    None of which is to criticize everybody who’s been doing metacommunity ecology for the last decade! Variance partitioning was a creative idea well worth pursuing. But I agree that it’s time to move on. And I think that’s happening. At least, some commenters on that old post of mine (and some email correspondents) say that it’s happening. And Logue et al., which Will linked to, picks up on some of these issues as well.

    • will.pearse says:

      I suppose I’m always hopeful that, as computing power increases, we’ll be able to fit explicit models more easily to data, and so methods like variance partitioning that are easier to fit won’t be our only option. At the same time, I feel like I’ve been writing sentences like the above for many of the last few years, so maybe that’s not the only thing holding us back.

      Anyway, I like the clonal analogy in your linked post. I think there’s a lot of overlap between fields that has yet to be tapped; I feel like Lynsey’s making good head-way doing so at the moment though 😀

      • Jeremy Fox says:

        Maybe. I certainly agree that fitting process-based models to data would be an improvement over fitting purely statistical models and then trying to figure out the “mapping” from the underlying processes to the statistics.

        But it’s my impression that the whole history of work on, say, phylogenetic reconstructions of trait evolution is of people fitting a more complicated model, thereby revising previous conclusions–and then having that conclusion further revised or even reversed when somebody fits an even *more* complicated model. You and Lynsey surely know the literature *far* better than I do here, so maybe my impression is *totally* off base. And I certainly wouldn’t suggest that people should just throw up their hands and stop trying to fit models to data, that would be silly. But I’m not sure if this is one of those situations where more computing power, thereby enabling us to fit more complicated models, is necessarily helpful. In The Signal and the Noise, Nate Silver addresses this directly, suggesting that more computing power and the associated ability to fit or simulate ever more complicated models is only helpful in cases like weather forecasting, where we *know* the model and so are limited only by our ability to simulate it (e.g., simulate it at sufficiently-fine spatial resolution, in the case of weather forecasting). In contrast, Silver suggests that increasing computational power is of no help at all when we don’t know the right mechanistic model, as in the case of trying to model earthquake occurrences. If that’s right (and it might be wrong, though it seems plausible to me), then the question is, do we know what metacommunity model we’d ideally want to fit, given sufficient computing power?

      • will.pearse says:

        Well, I agree that fitting a too complex model, or a model where the parameters aren’t properly mapped onto things of interest, can cause problems. I guess what I’m hoping for is someone to take models where the parameters we actually care about (e.g., dispersal-rate as a function of distance, competition coefficients, environmental preferences) are predicted to have certain values under certain models. I feel that, as long as you’re careful about parameter estimates and confidences (which I think Silver would agree with) then you’ll be OK.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: