Complexity in experimental ecology
There are various dimensions of complexity in experimental approaches in the environmental sciences
Experiments in ecology and environmental science are a very important feature of our science, and complement observational approaches. Even though people typically envision something specific when they think of ecological experiments (or do they?), these experiments really come in various flavors and span quite a range from Petri dish studies to field experiments.
Within this range, there are various aspects of complexity. Many of these aspects are linked but some are not. I see the following facets of complexity:
Complexity in organismal representation. You can work on a single organism, let’s say a particular fungal isolate, you can work with pairs of interacting species, or you can work with entire communities of a group of organisms (like a fungal community). And finally, especially when you work on soil, you can work with a complex set of interacting species of multiple groups, like an entire food web, consisting of bacteria, fungi, archaea, protists, nematodes, microarthropods and so on.
Complexity of environmental representation (abiotic factors, spatial scale). It is possible to work in homogeneous environments that are fully controlled, like the medium in a Petri dish, and all the way to a field plot, in which there are complex environmental gradients (for example, vertical gradients in soil).
Complexity of environmental representation (abiotic settings, temporal scale). In parallel to spatial scale, experimental approaches can also encompass a range of different temporal complexities. For example, there can be highly variable factors, such as diurnal or seasonal temperature changes or precipitation events; or temperature and moisture settings can be held rather constant.
Complexity of cases represented (degree of experimental unit complexity). A key aspect of study external validity (or the degree to which results from a study can be generalized) is how many different cases are captured in an experiment. This for example could mean including different soils, plant species, or in the case of distributed experiments (like NutNet or DroughtNet), different field sites in different kinds of ecosystems all over the world.
Complexity of process representation (response variable complexity). Partially going hand in hand with the selection of the target species and the environmental parameters, you can also choose the degree to which different processes are represented in your experiment, or at least which ones are captured. For example, when measuring responses you can measure just one (like biomass), or you can measure a whole range, and then aggregate these measurements into multi-functionality.
Complexity of factors (treatment complexity). One aspect of experiments that has been heavily under-explored is the complexity of factors examined. In a systematic mapping of the literature, we earlier showed that 98% of papers examining the effects of global change factors on soils in experiments looked at just 1 or 2 factors. This is in stark contrast to the reality that at any given time many factors are active, influencing ecological systems at any scale.
What level(s) of complexity you choose depends on your question, of course. If you are more interested in dissecting certain mechanisms at the organism level, you would try to keep other conditions simpler, for example. Conversely, if your goal is to capture as many aspects of ecological reality, your study would try to capture and represent as much complexity as possible.
I think it is interesting and important to realize just how many dimensions to complexity there are in experiments in ecology. This is not just important if you design experiments, but also crucial to keep in mind if you use data from them and have to interpret the findings. The complexity maximum is typically achieved by going to the field, which is often challenging. But what is the minimum complexity for addressing any given question?
Given how much experimental ecology has embraced different dimensions of complexity, it is perhaps surprising that the factor complexity is so poorly developed. This is due to a number of reasons, the most important one being the combinatorial explosion problem: the number of treatment combinations increases rapidly with the number of factors included. In an earlier newsletter I explain some of our current research on this, which involves asking questions about the number of factors instead.
Did I forget any dimensions of experimental complexity? Is this a pretty unique feature of environmental science, or is this perhaps similar in other fields as well?