Using combinatorial experimental approaches to discover better strategies in ecological restoration

Can we discover better ways to restore ecosystems by using combinatorial experimental approaches?

Combinatorial approaches to ecosystem restoration (AI generated image from Bing Copilot)

We are currently in the UN Decade on Ecosystem Restoration, and restoring ecosystems is something that we as a research lab have been interested in for a while, even though we do not directly work on this topic. But we do work on topics that have a lot to do with restoration, like soil, mycorrhizal fungi and biodiversity. And also global environmental change.

In our work on global environmental change, during the last few years, we have been carrying out experiments on a large number of factors, like 10 or more factors acting at the same time. You can’t do full factorial experiments, because this would be too many treatment combinations, so we instead have been working with a ‘random sampling from a pool of factors’ approach: we randomly pick global change factors from a factor pool and create replicates for a gradient in factor number this way. It is a bit abstract as far as experimental designs go, but it is effective — and we could show that the number of factors can predict some of the outcomes.

More recently, we have been thinking about flipping this question around, in a way, by asking if we can affect positive change in ecosystems by combining an increasing number of beneficial factors (rather than studying the mostly negative influences of global environmental change). Just like in global change biology, most experimental work in restoration has been done using 1 or 2 factors. This makes sense, because working with more factors is challenging and logistically difficult.

But what if by combining more factors, in this case management practices or restoration interventions, would lead to dramatically improved outcomes in ecosystem restoration? Wouldn’t that be worth knowing? We think so. In our first work on this topic, we introduce the general concept.

Just to make sure: this random sampling from a pool of management practices would mean that every replicate of the treatment level “2 management practices” would, by chance, be a different combination of two factors chosen at random from a pool of such strategies, maybe 10 or 20, or however many are feasible. The one thing all replicates have in common: it’s two management practices. And then we could do the same with 4, 6, 8 or 10 factors. Very likely, if there are positive effects of combining more treatments, this will plateau. So maybe from 4 or 6 treatment combinations combined, there are no additional positive effects on restoration outcomes any more. This would mean that this level could be optimal. But we don’t know if this will happen, since nobody has done this sort of work.

So this is what we think the process overall would be like:

• The first step would be to assemble a pool of possible factors (here: management strategies or restoration interventions). This pool could contain many items, more than 10, for example. The choices should be made together with interested parties, so that the factor pool is relevant to begin with, for example for a given world region or ecosystem type.

• The next step would be to start experimenting, maybe initially in the lab and the greenhouse, and to start trials using this random sampling from a pool approach. Another point worth mentioning is that while we advocate random draws of combinations as the basis for experiments, i.e. giving each restoration practice the same probability of being drawn, there may be good reason to deviate from this. For example, when practitioners based on their experience know certain treatments (or combinations) to be impossible or not ideal, they could be given lower priority, and vice versa. This means, while still essentially random draws, the different management practices could be given different ‘weights’ to reflect previous experience; this could make the whole process more effective. We envision such experiments to happen at regional restoration hubs, kind of modeled after agricultural experiment stations.

• Trials, eventually field experiments using small plots (this would not work in all ecosystem types), could then reveal unusually effective, and previously unknown combinations of treatments or a level of treatment combinations (factor number). These bits of information could then be the basis for recommendations for restoration practitioners; in the end, of course also logistic and economic aspects need to be part of the equation.

• This approach works best if there is broad participation and buy-in, so that we rapidly gain the knowledge needed to inform real management approaches.

This is the basic idea. Of course, realizing it will be a logistical challenge, but also one of convincing people to embrace this unusual experimental design with random draws.

We’ve written up this idea, and it is published as a preprint, and also accepted for publication in a journal.

How could we achieve that? What do you think?