Transcription
Good morning. So, my name is Vicki Templeton. I'm a professor of ecosystem functioning and services in The Institute of Ecology and the Faculty of Sustainability. And today, I'm going to be giving you a lecture on the ecological experiment. And I'm going to start principles, first principles, of what are the really important factors to consider when you're designing an experiment in ecology. Also in biology, these principles apply just as much. After this series of about 20, 30 slides on the principles, and then get to go into a whole load of different examples from other people's work and my own experimental work. So, I'm just going to be saying hello now with the film and otherwise going off so you can just focus on the slides and you can hear my voice. Yes, so we'll continue now.
Ecology is all about interactions between organisms, living organisms, and that monks living organisms, and the environment. And in ecology, we're very interested in very big questions about biological diversity and organisms and living life on Earth. So we ask questions like: why are there so many species, and how many, how come so many species can coexist? Why do we find these species in this location and very different species in another region or another location?
So often we're faced with something like this image: a landscape, a whole region in the whole mountainous landscape here. And on this image, you can kind of get a feel for the importance of scale and for inquiry and ecology, for answering questions that we have. Because on this image, you can see in the foreground there's sort of some gravel, uh, stony substrate in the mountain, and there are some plants, but not many. And then if you look further into the background, you can see a whole forested area going up one of the mountains. And to the left, you can see a mixture of forest and grassland. And then higher up the mountains, as you can see above the tree line, there is no evidence of plant growth. But probably if you went over there to that far off mountain in the distance, you might see similar patterns of small, low-lying Alpine plants that you can find here in the foreground. So you can kind of get the idea of this is really a question of scale. As soon as we talk about anything in ecology—who is growing where, who is doing what, who is interacting with whom—we immediately get confronted with this question of scale.
And if we want to understand this landscape and try and understand how does nature work, while we're finding these patterns that we're finding, then again we need to think about what might be causing those patterns. And this sounds like a simple endeavor, but it is certainly not.
And so what we often do is large scales, like say talking about this whole mountain range, um, we often use observational studies where we go and see what organs we find where. We come up with patterns, and then we also, we measure both the biotic components, so which organisms are there and what patterns, what kind of population sizes that do they have, how are they arranged, especially if we also then measure abiotic conditions such as climatic conditions, weather, pH, soil fertility, um, solar radiation, temperature, all kinds of different abiotic conditions, um, soil nutrients. We often measure and then we try and correlate the abiotic conditions with the biotic conditions.
And this then we can count with patterns and trying to work out what might be causing one effect that we're seeing on one pattern that we're finding. But in such a large scale, it's often a correlative affair where we don't really know. There are so many factors changing at the same time. For example, the weather over in the, in that grassland there, will probably be some management of, basically, there will be possibly some grazing animals put there by humans, or not. Nowadays, mainly yes, by humans. The forest will possibly also be managed. It says management factors, environmental factors, climatic factors, and also human factors of disturbance, the components like that. So you can kind of get the idea that there are many, many factors working at the same time and influencing what we find in this landscape.
So this lecture is about the ecological experiments. And these ecological experiments are probably inherently slightly different to other experiments that you will be probably hearing about in lectures from, uh, in other lectures in this course or other courses that you take. An ecological experiment, you'll be finding out later, has some really core components that need to be true for it to really be called an experiment.
For example, on this image here, you can see in the top right the Biosphere 2 in Arizona. And if you haven't heard of this, it's basically an enclosed greenhouse environment that scientists set up, um, with a full ecosystem inside that was set up with the aim to keep humans inside as well as the rest of the organisms and see whether they could survive in this enclosed environment, um, however the ecosystem would function properly and whether it could be a self-contained unit, obviously with the view to trying this out in space at some point. And this Biosphere project is pretty famous for not having worked in the end for various reasons I'm not going to go into today. You can go and look it up in your own time, um, but essentially, yeah, the non-linear dynamics of such ecosystems meant that they couldn't really predict from putting in the different components of the ecosystem whether the whole thing would function well together, and it didn't in the end.
And the point I'm making here is that some people call that an experiment, and it is an experiment in the sense that you're testing something, you're seeing whether something works. But in an ecological sense, if it were really an ecological, strong ecological experiment, you would have to replicate. You would have to do several biospheres. Well, somehow within that whole container, you'd have to have several ecosystems within your big greenhouse. Then you could technically call that an ecological experiment. So that's an important point for you to take with you first, is that we'll be hearing about this later, don't worry about it, but replication and randomization and appropriate treatments and meaningful treatments and appropriate controls are at the core center of ecology.
So we'll be talking about this quite a lot in this lecture. I've shown you some other pictures here of experiments that I'd been involved in. One on the bottom left as a close-up of a habitat assembly grassland experiment I set up a while ago in Euless where I used to work, but we really were looking at plant-plant interactions and also looking underground, as you can see in this picture. You can also then see, um, in the event experiment in Byroid where I also worked for as a collaboration partner, um, but we often as ecologists set up different plots. And in these different plots, these small, in this case, square areas, we apply a certain treatment. Certain kinds of organisms, certain kinds of abiotic conditions are simulated. So in this image, you can see a student adding rain and simulating heavy rain events to some of the plots. The other plots you can see have rain out shelters, and there they're applying a drought treatment. You see what we often apply factors to then see what effect they have. So we're looking at cause and effect with these experiments.
So you will probably have already guessed, but the answer to the question, why do we love doing experiments in ecology, is that we can actively manipulate specific factors. So if you think about back to the picture of the mountain, trying to understand what is behind all those different patterns that we can find in a mountain range is extremely difficult, um, it's possible, but it's extremely difficult to get to some of the more mechanistic details of what's behind what you find without actually carrying out maybe complementary experiments. So the experimental approach provides more reliable information on cause and effect. And I've added you a very nice quote from the Carbon and Hunsinger 'How to Do Ecology' book that quite a few of these slides are based on. And the quote says, 'Understanding cause and effect is critical, powerful, and much more difficult than it sounds.' So yes, we love doing experiments in ecology. I'm an experimental ecologist myself, which is probably going to shine through in this lecture, and, um, but it's nothing easy affair, but it's a still a very powerful and important thing to do. This is why we continue to do them.
So there are two types of ecological or biological data: observational data, which often comes in the form of either presence/absence (is this species there or not, one zero), counts (how many of whatever you're counting), biomass (for example, for plants, less over animals), um, and it basically involves what people, scientists, observe and can measure in the field, so out there in nature, in the landscape.
And what happens with this data then is that we try to make logical connections between the different components, as I said about the mountain range. We try and correlate say weather conditions, soil pH with whatever plants we're finding in a forest or in a grassland, for example. The weakness with this is that it's quite tricky, as we'll be talking about in a few slides, to infer causation from correlation. So just because two parameters or two things you can measure correlate with each other, they change similarly as one gets bigger, the other one gets bigger, or the other way around, doesn't mean to say that one is causing the other.
Whereas in experiments, we actually actively manipulate the environment to test effects of factors on parameters. So parameters are things that we measure, and by doing this, we can actually test effect sizes of factors. So what does this actually mean in normal, normal language? It means that we can look at the relative importance of a factor compared to say another factor. So we can have different treatments, so factor one with say a high or low fertilization, factor two could be a drought treatment or heavy rain as in the event experiment, and then we see how those two factors change say species diversity of a grassland, and then we can kind of see which one has the biggest effect, and we can also look at the interaction effects between those two factors.
Importantly, natural systems are naturally variable, which also makes it difficult for us to detect a signal of a certain, of whatever we're interested in, from the background noise, so to speak. So there's a lot of variability, and if we test a factor in an experiment in ecology, we're trying to derive the signal from that factor that we've manipulated and distinguish it from the noise. The noise would be that we call it residual variation, the variation due to other components, not the fact that you're testing.
So what are the strengths of the two different approaches? Well, for the observational, the great thing here is, of course, you're not really limited by scale as much. So you can work at a wide variety of scales. You can have larger areas. Um, you can do experiments, so to speak, in inverted commas, um, for example, you could go across that mountain range and say in different grasslands you could test a certain hypothesis and then try and come up with the reason why you're finding slightly different patterns in different grasslands. That's a kind of experiment, but it's not the same as the classic experimental one that I'm talking about on the right.
For the observational data, generally you have longer time frames, the largest time, larger time frames, larger, um, spatial scales, and very importantly, some things that we measure are only visible, they only emerge to have an emergent property at larger scales. Equally, you can also say that only certain things only emerge at smaller scales. So scale actually, going to take home as a key message from this lecture, is at the core of ecology.
And for these observational studies, um, the nice thing is that you have something realistic at the whole landscape scale. So whatever you're looking at, the interactions, the occurrences, they are happening at a pretty realistic scale. Um, this is why often when we talk about the experimental approach that I'm talking about next, people question how realistic it is. On the other hand, you could also say that the observational is not so realistic because you don't know what's causing the patterns you're finding. So you can criticize both approaches either way.
In the experimental, because we're manipulating factors and then seeing what effect that has on the response variables, so what we're measuring are the response variables, we can actually to some extent infer cause and effect, and we can try and tease out different effect sizes, effectors, like I said before. So we can actually control certain variables that we're interested in, like fertilizer, drought, rain, number of species, kinds of species, things like that. So we do controlled experiments at different levels of control, and this can range from a larger scale field scale, which you saw in the Bible experiment, you can go down to greenhouse or mesocosm experiments outside. We also have climate chambers that we work with in my group too, where we can adjust the climatic conditions and then have light and the climate change and grow plants and see, apply other treatments to them, and then see how they respond to these climatic conditions with the other factors. And you can go even smaller down to a really tiny within-organism physiological experiments.
So what are the limitations of the two different approaches? For the observational data, correlations do not infer cause and effect. And as I said for the mountain range, many factors are working simultaneously on organisms in this natural landscape setting. So you never quite know what is causing what you're seeing or what you're measuring.
In contrast, with experimental data, it's limited by the imagination, the intuition, and the experience of the experimenter. So whatever he or she comes up with, thinks of, and knows about the reality of the environment, like what is a realistic, meaningful treatment. An example would be fertilizer: if you're going to do high and a low fertilizer treatment for some plants, if you don't know what constitutes a normal range of fertilizers that a farmer might apply or that plants might experience in nature, you might go for some value that's gonna have a huge effect, but it's totally meaningless because, um, it's not anything the plant would ever experience in real life, either on a farm or in nature.
Then for experimental data, there's the issue of scale. We often, you know, experiments take up, uh, they need a lot of men and women power to set up and to maintain. We need money to fund, to keep funding the management and the upkeep of these experiments. We need people power to really go and take the, make the measurements, to analyze the data, to write up the stories, um, so we only can do experiments up to a certain size depending on our resources of human power and, um, money as well, space.
So then is the issue of realism. If I, you know, if I find really nice results in my, in say, the event extreme weather events experiment in Byroid, how much can we scale this up? Is this going to be realistic for larger scales? Is often the question.
So by now it should be clear that scale is really at the core center of ecology, and it plays a central role in determining the outcomes of observations that we make. And I really like this quote: 'The scale acts in what Hutchinson, who's a famous ecologist from last century, has called the ecological theater, and these acts are playing out on various scales of space and time. To understand these dramas, we must view them on the appropriate scale.'
I'd like to add to that that we don't just need the appropriate skill, when you can look at them at different scales and then bring these different outcomes that we find together into integrated knowledge.
And from a, from a point of view of talking about ecosystems, which we do a lot in ecology, ecosystems are systems where we consider both the biotic components and the abiotic components. Then ecosystems range in size as well enormously. So here scale is incredibly important, and we can have ecosystems that are tiny to enormous ecosystems such as the global ecosystem that you can see at the top here.
So we can go down from top to bottom here from the global ecosystem down to say a watershed, a whole area that is influenced by a river, down to a forest ecosystem here at the bottom. We can even study an ecosystem here called the endolithic ecosystem, which is, um, the rock surface, so between rock surface and roots and soil. All of these can be seen as different ecosystems, which really illustrates how scale is important, and we need to define our scale. And again, this is quite amazing because there are 10 orders of magnitude difference in size between these different ecosystems, but they all are called an ecosystem. So it's very important what we define, that we define this level of scale that what we're working at when we're talking about things in ecology.
And if we're interested in patterns in space, like in this colored image here, then you can separate the space into different pixels or grains, which would be like the actual, the cell, the minimum cell that you're looking at. And then you can see different patterns: some areas are green, some are brown, some are yellow, some orange, and we're trying to understand why or how these things differ over space. Now there's two very important components of scale, and we're thinking like this. One is the grain, and you can see this here depicted on the top right. This little square, one pixel, is the minimum resolution of the data defined by the cell or minimum polygon size. So if we're looking at a smaller scale and in landscape, it would be a plot. If we're thinking about satellite imagery, it's really going to be your, your pixel on your image. And then there's the other part, it says the grain is the smallest unit of your data, and then the whole unit that you're looking at is the extent. So this is the scope or domain of the data. So it's the size of the landscape or study area and the consideration. And this is very important in ecology, and it's a very difficult one to sort of pin down the extent because often we talk about whole ecosystems, like I did in the slide before, um, and the, you can have very large and long philosophical discussions about where there's an ecosystem end and where does it start, where does the landscape end, where does it start? Essentially, there's not really much point in arguing about this for too long. The main point is that you kind of define your area and say where the limits are, and then just describe what your green is. So you have to describe and define your extent and your grain in whatever study you're working on.
As I said, I'm an experimental plant ecologist, and for me, I really enjoy working at different scales because we can gain different information from the experiments that we do at different scales.
So if we start here, here's some images of experiments I've worked in, mostly all of them except the one the top left. So here in the bottom left, you have this riser box where we grew two Brazilian tree individuals in classic Brazilian soil. And with these riser boxes with plexiglass on one side, we can track the root growth, and in this image, you can see that they send the roots straight down the taproot without endearing to interact with each other.
And then we often look at interactions between plants at a slightly larger scale here on the greenhouse table where you have different species in a pot and more than one, more than two species or more than two individuals per pot, so it's more like a natural community.
You can go then to a larger scale still in the controlled experience. This is a large greenhouse in Eulish and the Flat Science Center where I used to work. They have these huge rhizotrans there that could get moved around by robots and go into a measuring chamber and get photographed like Hollywood for plants. Um, this is a really amazing scale for what we call phenotyping, where you want to measure certain traits of plants again and again over time on an almost industrial scale.
Um, and then intermediately between the controlled experiments and the field experiment is something you can see on the right. So you have this image of a little hut and this funny structure here, um, and this is an experiment I did in Euless some years ago where really had quite technologically advanced measuring equipment, um, that could move over the small plots and take pictures again and again over time. But this was still quite small scale compared to these field experiments that you can see at the top.
So I'll just finish what I was saying on the last slide, um, so you can see the big field experiments on the previous slide. And there now we're really talking about a number of hectares for one full experiment. What you can see now this slide is an intermediate thing which is very, very useful approach that we use a lot in my group, and it's called mesocosm. So what we do is we combine the kind of a field experiment with a controlled experiment. So we have large pots, but we place them outside, so they're not in the greenhouse where we can control the weather or keep the weather out, but they're outside exposed to all the normal conditions that the other plants and the surroundings are exposed to. This is an area near the, near the quarters fed, it's the greenhouse area that we have in Lotusfield, you know, a little book. And here you can see this is the preparation of an experiment. We can manipulate different plant species and diversity levels. We can add species different times. We can collect leachate, find out what's happening, so the nutrients in the soil. Can do all kinds of more controlled things here, but it's not a very large plot scale, it's not a field experiment, but it is a mesocosm. And so we can kind of have the advantages of both the ex, controlled experiment with the more realistic weather conditions outside.
So in summary, I've provided you with this very nice picture from the 'How to Do Ecology' book, um, talking about both spatial and temporal scale and how to what extent we tend to do experiments, the different low spatial to high spatial, low temporal to high temporal scales, and to what extent we do observational studies.
And you can kind of see if you're talking about this scale down here of a, we're really talking about less than a meter and less than a minute, that's very, very likely going to be an experiment that we're looking at. Uh, if we're looking at B, if we're talking about spectral scale about 100 meters or less and time scale of about hours, this is still going to tend to be an experiment. If we go up to C and D and E, then we're really talking about larger scales, but particularly much longer time scales, then this could be anything. It could be an experiment, but it could also be an observational study. And often when we get to this very large scale, very long term, then we very often end up doing the observational studies.
And the important thing is not that one is better than the other, but the combination of the all these different approaches is very, very important for understanding our systems. What we also do increasingly now is we, we use modeling, computer modeling, to work out connections between these different scales and also to scale up from the smaller to the longer time and longer spatial scale.
So to sort of sum up what we've just been saying, manipulative experiments can help to establish causality in ecology. And I like this quote too because it's not like they're absolutely going to show you exactly what is causing the effect that you're finding in an experiment, but experiments do help. And again, it's this saying of standing on the shoulders of giants. This is what we do in science: we, we add a little bit of knowledge to what's already there on that giant, and then the giant gets bigger. And so both these approaches are equally valid and very important. And I'm not going to go into all the details, you can read them again, what the differences are between the observational and the experimental studies. Generally speaking, as you can see from the slide before, the observation study is going to be a larger scale, but you can also do experiments at larger scales, which is increasingly happening across the globe with these networks of scientists doing the same experiment in different locations, um, and experimental studies tend to be the smaller scale, but as I said, nowadays there's these things like nutrient net, drought net, all kinds of different networks of scientists, so they end up doing quite an amazing replicated experiments across different regions of the world and over time.
So before we start talking about the really important components of an experiment, if you want to decide, design an ecological experiment, I just want to give you a few words about correlations and cause and effect. So it's kind of a word of warning. So there's a collective wisdom that correlation does not imply causation, whereas Bill Shipley has a much more nuanced and I think excellent way of describing it, and he says, 'Correlation almost always implies causation, but cannot resolve which of the two variables caused the other.' So think about that very well because it's a really nice sentence that is extremely true that we often if we see two things that correlate to each other, then we kind of have this feeling, we, we know that there's some kind of implication that one caused the other, but we don't actually know, we can't find out which one caused the other. So when we do multivariate statistics, which we'll no doubt meet during the course of your degree, um, this is based on correlations, so bear in mind that it yes tells you what's connected, what's correlated to what, um, but it doesn't necessarily mean that you're going to know what caused what. But over time, if you do enough studies, you might get an inkling, and you could then go and do an experiment to test your hypothesis based on your correlations.
A classic example of this is the saying or the feeling people have that red cars have my accidents, or actually it's not just a feeling, people, there are statistics showing that red cars are often more involved in accidents than other colored cars.
And so this student, Rick, at the end of graduate school in USA, um, he had a very old car, and his girlfriend convinces him to buy a new one before starting his new job. And Rick's favorite color is red, so Rick was thinking of getting a new red car, and his girlfriend is against this. She says, 'Don't do that because red cars are involved in more accidents per mile than other color cars.' So she isn't playing that red is a cause of danger and that he would then possibly have an accident because he's going to buy a red car.
What's the problem with this logic? I'm sure you can probably work it out, but basically the model of reality here is either the one on the left, which is the red is on the excess is causing danger on the y-axis. So in science, we often have the explanatory variable on the x-axis and then the response variable on the y-axis.
But equally, it's possible that, you know, dangerous or danger, dangerous behavior somehow connected to red cars, and this isn't clear from this statistic saying that, you know, more red cars or red cars are more often involved in accidents than other colored cars. So in essence, this hypothesis saying, um, red cars are a cause of danger, it could be wrong. It could be the other way around. It could be that danger, dangerous behavior is correlated with red cars. And this sense, it possibly is really the explanation for this statistic that people find is that people who tend to drive dangerously like red cars. They like sports cars, probably, and they like red cars.
Here I'd like to provide you with a more academic, scientific example of the issue of correlation not knowing, um, and not knowing what is causing what. And this is from the plant stress insect herbivore hypothesis. And this is something that, um, scientists got, White, worked on a lot in the 80s and 90s. And it comes from this idea that, um, people noticed, scientists noticed that when plants are stressed, for example, by drought, and they have water stress, then they often get attacked more often by invertebrate herbivores. So the herbivores are somehow attracted to these wilting, not well water supply plants. And the hypothesis behind it that the scientists came up with is that this stress, because they have less water availability, the plants, it means that the nitrogen in the leaves is more concentrated. So there's an increase in the amount of nitrogen available in their tissues, and the herbivores love nitrogen, it's protein source for them. So therefore, they'll particularly go for those kind of leaves, and this can cause outbreaks of such phytophagus or plant-eating invertebrates.
So that's the hypothesis. Um, here in the image, you can see in the mobilization during a leaf senescence, so M moves around the plant over the spam of the plant's life depending on the seasons, but that's not really highly relevant for this right now.
So maybe you've noticed this yourself before on plants. Maybe you know something you have on your own house plants. The water stressed plants tend to get attacked more often by insect herbivores. So here you have some green fly, I'm not quite sure what's happening here, but it's definitely wilted and attack by something. Here you have Visarium fungus attacking these cabbages, and here you have water stress, and I don't know exactly what's attacking it, but it's basically different examples of wilted, stressed out, water stress plants being attacked by insect herbivores.
The question behind it, though, is what's causing the poor health of the plants or the poor performance? Is it the water stress or is the insect herbitals? Or why, why is this correlation there? I mean, what comes first? Is it water stress or the herbivores? And of course, the more logical hypothesis is that the water stress, um, means there's less water in the plant, and therefore any nutrients in the plant will be concentrated, and the herbivores particularly like this because they like the nitrogen, the protein, they find it yummy.
So again, with the big key question of what is causing what, um, White particularly focused on outbreak of herbivorous sap-feeding psyllid insects in relation to stress in plants. And you noticed that after this happened very often after wet winters and dry summers. So you note that there was a correlation with specific weather events or weather patterns. And his model then, or the reality that he was hypothesizing is what's happening in nature, was that you get unusual weather, then you get physiological stress in the plants, and then you get an insect outbreak. And the mechanism he was inferring was that stress increases the ability, availability of limiting nitrogen to the insects, so they then get attracted to those particularly water stress plants. So their model and a reality is weather stress, increased nitrogen, outbreak. So he elaborated on the original hypothesis of unusual weather, physiological stress, insect outbreak by adding this proposed mechanism of the increased nitrogen in the tissues of the plant because of the reduced water. Of course, there could be an alternative hypothesis that, um, plants that are attacked by herbivores get really stressed, and therefore they're not able to take up water as well, and so they might wilt and experience drought stress because of the attack by the herbivores. So then you don't really know which one is causing the other one.
The really nice thing the science though is that it's something that we often call standing on the shoulders of giants. And so we often work in large teams across the globe on different issues, and then we add our little bit of knowledge to the huge giant of previous knowledge, and so the power is when you keep doing different experiments that are slightly, slightly varied, and then you can find out well to what extent do you keep finding similar results, to what extent do they confirm each other?
Let's move to number two. A very important thing in experiments in ecology is having a control, to have something like a reference that you can compare your treatment effect to. So you compare your manipulation with a control. Now might sound trivial, but why, why do we need this control? Why can't we just see the effect say of removing the herbivores? Well, I can't really have a live lecture with you right now where we could discuss this, but if you think about it for a minute, you might come up with observation say in nature like do, do things stay massively stable all the time? No, actually in nature they don't, um, because biological systems often change over time. So they change naturally, especially if you, um, followers send you how it changes over time, then slowly it will get, um, have plants coming in that establish themselves, and they will stabilize the Dune, and then other plants that are less adapted to the totally shifting new June sand will be able to then come in and establish, and so on, so on, so it goes through succession and end up being a grassland over time, which could turn into a shrub land and then into a forest. That's called ecological succession. So we have this kind of change going on in the background. So if we're going to manipulate something and add a treatment in an experiment, how do we know without control whether what we've measured after we've applied the treatment is really due to this treatment or is it just due to natural background change? For example, you might provide some male deer with a strange diet, and then you might observe that they've shed their antlers in the spring, and you might very incorrectly conclude that they're shedding their antlers due to the static diet when actually all male deer shed their antlers in Spring. So you actually have to know your system, and you have to know what kind of natural changes go on in the background. So without this control, you wouldn't be able to infer causal effect of the diet.
Another nice side warning here is you need to find an appropriate control, but be aware, it's not always just no treatment. It's not always just no cage, or if you're sowing plant species, no, no sowing of plant species. It depends on the question. Small trees were exposed to elevated CO2 or ambient CO2, so the current concentration in the atmosphere at the time versus double that to simulate climate change effects of elevated CO2. And in this experiment, you can see here, it's long gone in Scotland, this is in Perthshire, about one hour north of Edinburgh.
And we exposed these trees in these chambers, the open top chambers, to either elevated CO2 or ambient. And as you can see, there's a whole load of different chambers on the left side. And here in this at the back top right, you can see there's some trees, um, but there aren't any chambers anymore. Maybe you can start wondering why, what's going on in that top right corner that's different to the part of the experiment where we have chambers?
Well, you can see on this map of this of the experimental design done by the statistician that say bottom left here you have an ambient chamber with high and low nutrients, and then you have a elevator CO2 plus CO2 ambient CO2 are randomly located across the side. And over here this top left I was talking about, um, you have the similar treatment, you have high and low, but you don't have a chamber. So essentially this is your control. This is one kind of extra control. So in this design, the elevator CO2 is the treatment. You want to see what happens to the trees when you double the CO2 in the atmosphere, so as it's happening right now as we speak, CO2 is going up at a fast rate, um, compared to the ambient. So that's your first control, which is what's the effect of doubling the CO2 compared to what is in the atmosphere at that time, at the moment, in this case for the experiment.
But these chambers, if we go back, if you think about those open top chambers, they don't just change the CO2 level, do they?
Change the light coming into the area; they change the humidity, so they have unwanted side effects. And this is why we have this extra control here on the top right with no chamber, but the same trees and the same nutrient treatment, and also ambient CO2 levels, because that was what was in the atmosphere at the time. So you can see, it's not always just no CO2 on, on current CO2. You also have to try and measure the effect of this, this artifact effect of the chamber.
As you can see here, coming to the third very, very important aspect of designing experiments, it's called replication. Now, what we need for experiments, for them to be powerful and strong, is we need independent replication of the experimental units of each treatment and the control. And why do we need this? Because we want to really separate the real effects of the treatment from chance effects that we've talked about, or background noise.
So, as we've heard before for the controls, biological systems tend to change. So there's some kind of background change, but it's not exactly what you're really focusing on; that's not what you're interested in, even though it's important. And sometimes there are chance effects; there are differences across sites.
So, if you just had your, for example, in my elevated CO2 and trees experiment from my PhD, if we just had one chamber with elevated CO2, a big one, pump it with, say, extra CO2 gas, and then we have another big chamber with ambient CO2, what would be the problem with this? Why do you think this would not be adequate? You'd only have n equals one; that's your replication. Essentially, if you found differences, you wouldn't know for sure if they were due to experimental treatment, or differences in local conditions.
So, let's say you put this big chamber on the left side of that experimental area and your ambient chamber on the right side. Maybe the right side is more shaded, or maybe the soil nutrients are richer on that side. So any effect that you might find could just simply be due to the local conditions and not your actual treatment.
Worst case scenario really is n equals one. So it doesn't matter if you had this one big chamber, even if you sampled inside that chamber again and again, and you have a high precision of sub-sampling, it doesn't increase the replication. Your unit of treatment, your open top chamber, is only one: one elevated, one ambient in the example I'm giving you. Because the factors that affect one sub-sample may also affect another, so you'd have no way of knowing if your effects, the effects that you measure, are due to the treatment factors, or local conditions, or the individuals that you sampled.
A nice example from my previous workplace, in the Forschungszentrum Jülich Research Center in Jülich near Cologne, is my colleagues set up these chambers in the greenhouse. Whereby, with these blue constructions here, they could change the temperature of the roots in pots. So they can thereby sort of simulate what experiments are normally like for plants that are rooted in the soil, where the temperature of the roots is usually a bit lower than the temperature of the shoots. So they set up these chambers with these nice different lines of treatment areas. And then they could, say, put lower root temperatures in the one chamber on the right, and they can have higher root temperatures for there's an experimental treatment on the left, for one particular species or whatever plant species they want to have in their pots.
Now, what's the problem with this, if you think about what I just said? Essentially, you only have one treatment unit. You have one chamber with low root temperature and the other one with high. So again, it's pseudo-replicated. And this is a bit like this example in the picture.
So, if in this example somebody is looking at the effect of predators, a pot with plants that has herbivores in it, or whatever, it doesn't really matter. This person has put up a barrier in the middle of the table and then got nine replicates where they add predators, and then nine pots, or whatever they are, where they didn't add predators. Does this look good to you? It looks good at first sight, but it's actually pseudo-replication, because you really have all your non-predator treatments on one side and all your predator treatments on the other side. And so maybe there's more light on that side of the table, or there's more water on that side of the table, and therefore any effects that you find are really to do with the light and the water, and not to do with actually having a predator or not.
So, what should you do? It's a lot of work; doing good science is a lot of work. But essentially, you need to do something like this: you need to randomize and replicate. So, put all your replicates with P across the table in different locations. In this case, it's quite systematic, but it's actually random. And then put barriers around each one so they're not influencing each other, because of course you don't want a predator right next to a pot without predators; it can probably jump across. So this extra work is necessary to be really sure that, a) you have enough replicates, like you know that whatever happens in the top right of this table in that pot is one example, but you also have lots of pots on the other part of the table. And also, you can see that it's randomized, which will bring us to the next and final point about experimental design, and this is randomization and interspersion.
Do not underestimate the importance of this factor as well. So this replication, where you, you know, you want to find out, "Okay, does this just happen to this one individual, or does it happen to general individuals of that species?" It's very, very important. But it's only useful, the replication, if these reps are spaced correctly. So you kind of got the idea with the previous example, or the table with the predators and non-predator pots.
Another example could be you adding nitrogen fertilizer treatment to a swampy, wet area of grass, or you're adding to a grassland. Let's say adding high nitrogen and low nitrogen treatments. To make a link again to the meaningful treatments, to add a meaningful nitrogen fertilizer treatment, you also have to know, well, what is a kind of classic fertilizer level that a farmer or a landowner would use if you're interested in, say, growing crops? So you have to know what kind of level am I going to apply, like not too much and not too little, you know, you know any effect on the plants. And that's the side story. What I'm really wanting to emphasize here is that if you have the high N treatment and the low N treatment, if you just applied high N to one side of the grassland and low N on the other side, like in the previous example, then it could be that you have applied low N to the dryer area and by chance applied high N to a swampy, wet area. And so if all your replicates of one treatment are found on one side and the other treatment, the low N, is found on the other side, then you might actually be measuring the local effects of topography or conditions, and you've got no way of knowing if your treatments really cause the effect that you measure.
This is an example from a very large plant-soil feedback experiment that we did in Jülich a long time ago, called 'Home and Away' unit. We took a lot of soils from a biodiversity experiment, and then we planted, we basically planted the species that had been growing on those soils, or a new species that was either species was feeling at home in its old soil, or it was in some new soil that had been affected by previous species. You don't need to understand the details with all these different colors in the pots; there were a thousand pots on the tables. They denote different treatments that were then applied to these plant species. I wanted to find out what affects the total microbiome, all the microbes in that soil, whether it's negative or positive or neutral for the plants that are then grown on it.
What's the issue here with having a thousand pots on greenhouse tables? So think about what I said before. What did we have to do quite regularly? Because when we set up the experiment, for example, we wanted all the orange treatment, whatever that was, to be in one corner of the table, because then we could treat it in one way. And the same thing, say, for the green ones or the pink ones. But then, when we were actually set up the experiment, once we'd, you know, put the plants in and we wanted to just let the plants grow then and see what effect the treatments had, then we had to move them into a randomized, dispersed design. So that, for example, we didn't have, sort of, often there's more steam, or more moisture over here near the side of the greenhouse, and there's different light, or it's windier in the middle of the greenhouse. We didn't want an effect of like treatments on the side to affect a particular treatment more than others. So we had to intersperse them and make sure that we did this every few weeks. So the experiment lasted for six weeks, and then we had to really just randomize, move the pots around with all these student helpers and students into different locations.
What you can see now is what we did when we wanted to sort of apply certain treatments, say the pink on the green. And the pink was the control, and the gamma-irradiated, so we kill the microbes, is the green labeled pot. So you can see this is what we did when we actually wanted to set up or measure something: move them so they're all together. But then, of course, we want to allow them to just grow and be affected by the treatments, then they needed to be randomized again. So, of course, which picture is the best for making sure I needed units, sorry, are placed at random? Obviously, the pictures at the top, where especially this one, the top right, where they're nicely interspersed, they get moved around every few weeks.
So, I hope that this has given you a nice overview of how experiments are fantastic. They can really find out cause and effect, if you're lucky, but they require a lot of thought and preparation. And obviously, you have to work harder, maybe than you would have thought. Think about all those barriers that you need to put around the different predator/non-predator pots in the example I showed you on the table. There are four key issues that you need to think about if you want your experiment to be successful: You need the treatments to be meaningful but realistic, but also not have confounded side effects, or if they do, you need to measure them. You need appropriate controls, like a control that really is testing what happens if you do nothing. And again, think about the extra control example I had from my PhD, where they had the open top chamber with ambient CO2, but then you also had ambient CO2 with no chamber. You need independent replication, like they should not be dependent on each other, so not all in one part of the table, or not all in one part of the experimental field. You don't have one treatment on one side and the other treatment on the other, because you don't know what local factors might be playing a role. And very importantly, for the replication to have any power, you need the location of your experimental units to be randomized and interspersed over in space.
So, in summary for this part, [Music] this is actually a summary from with some other slides, which we can talk about another time. But basically, what's important to also think about with all these issues about designing experiments is, well, what is natural science based on? And I'm going to give you some extra reading on this, but it's based on hypothesis testing, as well as knowing your system. So to come up with meaningful treatments, you have to know your system. Like I said, you have to know what would be kind of a normal fertilizer level for this crop or whatever you're studying. If you throw on the double the amount, or only a fifth of the amount, it probably isn't going to have any biological effect. You also need to know about the null hypothesis, which you'll hear about later in other lectures, I'm sure, which is that there is no effect of your treatment. What we're trying to do with experiments is derive causality from our treatments that we set up. So we want to say and find out whether one factor causes an effect. And this is often different from correlation. Correlation is just when two different things you can measure behave in a similar way. So if you increase one, then the other one increases. It doesn't actually tell you which one caused the other; it just means that they correlate, they basically behave in a similar way.
Other things to think about when you're thinking about experiments is, what's the kind of scale your experiment is at? If we have quite small pots, like in some of my examples, you can really control things quite well, and you can do many, many different treatments, lots of replicates. Of course, this isn't as realistic as doing a field experiment; that's why we also go into the field. But even in the field, where we have these plots that could be a number of meters squared, this is still a smaller scale to what you'd find at the whole field level, what your farmer or a conservationist would be dealing with. And this is where we often go into modeling approaches, because it's just too much work to set up experiments where one plot is a whole field, although people are starting to do this.
So, in summary, if you recast your memory to what you've been learning so far designing experiments, there are four key components to it. What are these four components? Maybe you can just think through them in your head. So, to be able to infer causality, we need what kind of treatments? Meaningful treatments. What kind of controls? Appropriate controls. What kind of replication? Independent replication. And we need randomization and interspersion of treatments.