Often in ecology we cannot do the experiments we want due to logistical constraints and ethical concerns. This is especially true when studying species invasions. Ethically, we cannot release an invasive species in multiple locations to see what happens, and even if we could, the scale of species invasions (which typically happen over tens of thousand of km2) would make any experiment extremely difficult. As a result invasion ecologists have often relied on natural experiments, studying the behaviour of the many, many, populations humans have intentionally and unintentionally established outside their native range.
This approach has also been used to try and understand what makes a population go from naturalised and relatively harmless, to a rapidly expanding invasive one. The time this takes is called an invasion lag. Due to the importance of managing invasive populations when they are still small there is a lot of interest in predicting how long invasion lags will last. Studies into historical invasion lags have provided several interesting examples of populations becoming invasive after pollinators and seed dispersers were also introduced, hybridisation between species, and even as a result of war. However, we often don’t start paying attention to introduced species until after they become a problem, and so miss the crucial early spread. With few examples to draw on, and with a many possibly causes for the transition from benign to invasive, these studies are historical examples we fit a narrative to after the population has already become widespread. It is unclear if we can build a predictive model based on such evidence.
In our new paper we test how well a discrete event (a change in dispersal ability) could be linked to the timing of the invasion lag breaking we used a stochastic simulation of a plant population spreading over a fragmented landscape, where we change the dispersal ability after 50 years. Both the change in dispersal ability and gaps in the fragmented landscape cause invasion lags. Because we control the simulation we know the “truth”, and can compare multiple simulated invasion trajectories to this “truth”. In effect we simulate a simplified, controlled, replicated, invasion lag experiment that we cannot carry out in reality.
What we found was not encouraging for our ability to predict when invasions lags will break. In fragmented landscapes we saw long invasion lags even when the population had high dispersal ability the whole time. This was especially true when the invasion started on the edge of the habitat (Figure 1a). The effect of habitat fragmentation on invasion lags was large enough that even when 50% of the habitat was suitable (which did result in fairly connected habitats) the median length of invasion lag when dispersal ability changed was within the 95% bounds of invasion lags when the kernel did not change (blue bars overlap red circles Figure 1)
Figure 1. The effect of habitat fragmentation on invasion lags when dispersal ability is high the whole time (blue) and when dispersal ability starts low and increased after 50 time step. On the x-axis is the time step after introduction that the invasion lag broke (we estimated this using a moving average spread rate), on the y-axis the degree of habitat fragmentation. Dots show the median length of the invasion lag and bars show the range that 95% of simulations fell within (1000 simulations per bar).
Imagine if we only got to see one or two of these simulated invasions, could we tell if the dispersal ability had changed or not, and if so when it had. The answer, even in this very simplified system, seems to be no. Yet this is what is commonly done when reconstructing historical narratives around invasion lags. An event is picked as a potential trigger (in hindsight), and an increase in spread rate after the trigger event is taken as evidence. However, our simulations show even a single additional process introducing an invasion lag (in our case fragmented habitat) can produce enough variability in lag times to obscure the effect of a marked increase in dispersal ability and the timing of that ability.