We just had a new paper on community assembly concepts come out in Ecology Letters. The focus of the work is on predictability. Making community ecology useful for applications like conservation, land management, or agriculture requires the ability to predict the future rather than simply explain the past. Strong predictions are a key property of useful science, and mark the beginning of a transition away from discovering basics to engineering solutions. Much as we use physics to successfully launch satellites, we would like to be able to use community ecology to understand human gut microbiome, restore landscapes after disturbances, or control the spread of invasive species. These have all been challenging goals. But are they impossible? This paper asks when prediction is and is not possible, and highlights the range of complex scenarios that can occur in communities experiencing change.
A classic example to motivate the work comes from elevation gradients – here, a mountainside in northern Taiwan. As the climate warms, say, the simple expectation is that lower-elevation (warmer) communities will replace higher-elevation (colder) communities. But often this doesn’t happen. The paleo-record shows that when climates change, communities often lag in their responses to that change, and sometimes end up in wholly novel states instead. The contemporary record also shows that alternate community states are possible when disturbances occur. Any of these scenarios will yield a system with lower predictability, and less hope for us as community ecologists that we will be able to contribute to useful applications. Nevertheless many current ecological applications – correlative species distribution modeling, transfer-function based paleoclimate reconstruction, etc. assume that none of these effects are important.
In this paper we put together a framework that compares a commmunity’s state over time (C(t)) to an external forcing variable over time (F(t), e.g. climate change). If the system tracks this change and is predictable, then these two variables will be 1:1 linearly related – but they may not be. We develop a ‘community response diagram’, here shown in green, to highlight this potential divergence.
In the rest of the paper we demonstrate that in many cases, the community can show strong lags (divergences from the 1:1 line) and also alternate unstable states (multiple values of C(t) for a single value of F(t), shown here as a ‘n’ statistic indicated by the vertical black line) that reduce predictability. We also develop a differential equation model that explores the realistic community-scale processes that could generate such patterns.
The overall message of the paper is that predictability may be the exception rather than the rule in community dynamics. Using these community response diagrams helps us delineate when prediction is and is not going to be possible. We have carried out applications with real data, but were not able to include these in the final accepted manuscript – look out for them soon in a separate publication!