I’m about to start making a lot of posts about niche and distribution modeling as the project I’m working on for the California Department of Fish and Game nears completion. The point of this project is to build correlative models of species ecological tolerances and to use those to predict the effects of climate change on the distribution of suitable habitat for terrestrial vertebrate species that have been designated as “species of special concern”. I would really like for this work to be understandable by non-specialists as well as specialists, so I’m going to start by making some introductory posts that just explain what I’ve been doing and why.
In order to understand the effects of climate change on suitability of habitat for a species, we need to know what exactly causes habitat to be more or less suitable for that species. Does it like high temperatures or low temperatures? Is the maximum temperature more important than the average temperature? Does it like wet or dry environments? Can it tolerate higher temperatures when it’s wet than when it’s cold (i.e., are its tolerances for environmental factors correlated)? Which environmental factors are important and which are unimportant?
What we really want to know is a thing called the species’ fundamental niche. This is the range of environments within which that species can maintain a population that reproduces frequently enough that the population does not shrink. The fundamental niche does not consider interactions between species or the ability of species to disperse to different patches of habitat, it only considers what sets of environmental factors a species can tolerate. For the sake of illustration, we’re going to talk about the fundamental niche of a cartoon frog:
Here we see an environmental space in two variables: average temperature and annual precipitation. Within the set of environments circumscribed by the green ellipse, the frog is happy, so happy that it’s reproducing at a sufficient rate to persist. Outside of the green area the frog is not so happy. Far enough outside of the green ellipse, the frog is dead. In order to know whether a particular habitat is suitable for our frog, all we need to know is whether or not it falls inside that green ellipse. Easy, right?
The answer is “no, not at all”. It turns out to be a very difficult question.
One possible approach to answering this question is to grab a bunch of our frogs and bring them into the lab. We can then raise them in a range of temperatures, keeping everything else constant.
We see that our frog does okay when it’s not too hot or too cold, and we can get some idea of what the range of its temperature tolerances is. Then we can do the same for precipitation:
So what’s the problem with this physiological approach to estimating species tolerances? Well, it’s actually been a very productive area of research, but there are some issues that limit what we can do with this approach. One is that the resolution of the experiment is limited by the amount of animals you’re able to raise and your ability to maintain fine differences in the environments they experience. Despite my little cartoons here, you can’t really just do this with one frog in each box. You need quite a few so that you can average over all of the statistical noise that’s created by differences between individual frogs and the unavoidable effects of random chance. In the example above, let’s pretend that each one of those precipitation boxes corresponds to one more foot of rain per year. Zero and one foot per year are unsuitable, as is anything over seven. But how about six and a half? One and a half? We actually can’t tell where the cutoff is because the resolution is too low! Assuming we’ve got ten frogs per box, we’ve already used a hundred frogs only to find that we can only estimate their tolerances to the nearest foot.
There’s another problem, though. Remember our frog’s fundamental niche?
That ellipse has a tilt to it, such that our frogs can handle higher temperatures when there is more precipitation. By examining one environmental variable at a time, there is no way that we can detect this correlation. Essentially our study looks like the following figure:
We’ve fixed the precipitation at some value, and then we’ve seen that the frog can handle a range of temperatures given that value of precipitation. However, we can’t necessarily extrapolate that to the entire range of values that the frog could tolerate. If we had fixed the precipitation value lower, our frog would have been less able to tolerate high temperatures and more able to tolerate low temperatures than what we observed. If we had fixed the precipitation value higher, the opposite would have been true! In order to really understand the species’ tolerance for temperature, we have to examine it over a range of precipitations at the same time. Something like this:
Notice the big problem? With even the coarse resolution of our experiments (ten treatments per environmental axis), we’ve now got a hundred experiments to conduct! If we’re doing ten frogs per experiment, we now have a thousand frogs to take care of. The addition of more variables compounds the problem – with three variables we’d need a thousand treatments and ten thousand frogs, four variables requires ten thousand treatments and a hundred thousand frogs, etc. We’re going to run out of frog chow and research assistants fairly quickly at this rate.
So far we’ve been talking about frogs. Frogs are relatively easy to keep in boxes, but that’s not true of every animal. Picture the above discussion with elephants substituted for frogs and you quickly see that the enterprise is over before it begins – we simply don’t have enough elephant-sized aquaria to make even a low resolution study practical. There’s also the fact that these experiments may require more individuals than a natural population can spare – if you want to know the environmental tolerances of a species that is only represented by 500 surviving individuals, no governing body is going to give you license to catch half of the extant population and raise them in the lab, particularly in a set of experiments that may result in some of them dying or at the very least failing to reproduce to their maximum capacity. Finally, consider that some environmental variables that are relevant to a species may be very difficult to manipulate in a laboratory setting.
I don’t want to be misconstrued here – physiological studies of the niche such as those presented here (albeit in cartoon form) are among the most reliable methods for estimating a species’ environmental tolerances. The above is simply to point out that there are practical limitations to the resolution and complexity of the niche estimates that can be produced this way. The traditional infomercial approach at this point would have me saying “THERE’S GOT TO BE A BETTER WAY”, at which point I would expound on the glory that is niche modeling. However, I do not want to imply that niche modeling is a better approach than physiological studies – in many ways, it is significantly inferior. However, there are some limitations of physiological studies that niche modeling does not share, and there are times when the lack of those limitations is of great importance (e.g., you can easily build a niche model for elephants using a correlational approach, while a physiological approach would be quite difficult). Niche modeling does have its own limitations, however, and I’ll talk about some of those as we go on.
In my next post, I’ll talk about what the correlational niche modeling approach is, and how it overcomes some of the issues mentioned here. In later posts I’ll talk about a whole slew of new methodological issues that the correlational approach raises.