Managing threats to communities of declining species with incomplete information
By combining models of responses to threats with network analyses of species co-occurrence, we developed an approach to predict how an ecological community restructures under threat management
Information from a few species on co-occurrence and expected responses to alternative threat management actions can be used to train a response model for an entire community
This means incomplete data can still usefully inform our decisions around threat management
We know little about how ecological communities respond to the management of the multitude of threats impacting them. And that’s despite decades of studying species’ responses to environmental change and, more recently, the threatening processes that destroy their populations and habitats. Our lack of knowledge is driven partly by a failure to adequately monitor our management of impacts. However, more generally, it reflects our inability to collect enough information across space and time to build good predictions.
I recently went through my own research to prove this point. In 2016 I participated in three studies evaluating the effects of management on species in Australia. One was only able to predict the likely outcomes of protection mechanisms for less than 60% of nomadic Australian birds. Another study only predicted fire management outcomes for a mere 12% of Banksia species in south-western Australia. And the third managed to predict likely responses to threat management for 42% of woodland birds in the endangered Box-gum Grassy Woodlands of eastern Australia.
We are slowly learning the effects of some management actions on certain well-studied species. For example, in grazing or cropping landscapes where trees have been cleared, a number of woodland bird species such as the willie wagtail, grey fantail and many honeyeaters will most likely increase in a patch of remnant vegetation if we restore tree cover through revegetation or grazing exclusion. We also know that other species might decline if their patch of vegetation is changed through restoration, because they prefer the current open conditions to more closed canopy woodland (eg, the much-loved Australian magpie, and parrots such as the eastern rosella). However, for many species, we have no idea whether conservation actions, such as planting more trees, will cause an increase or a decrease in their population as they are poorly studied or hard to detect. Waiting for more information to be collected hampers our attempts to stop species declines and could even result in species’ extinctions if management is too late (or wrong).
We want to recover communities of declining species, but how can we do this if we don’t know which species will benefit from which management action? The answer is superficially simple:
collect more data to inform management. But in most cases this is impossible. We are constrained in everything we do by limited conservation funding and time. We simply cannot afford to wait until all the information we need is collected, as every day we wait pushes species closer to extinction (see Decision Point #60).
Filling the gap
Our solution to filling the knowledge gap is that we should predict responses to management of ‘unknown’ species by using partial data and coexistence theory. We hypothesised that species who co-occur in the presence of threatening processes would share a similar response to the mitigation or elimination of those threats, as niche differences allow them to partition resources and maintain co-occurrence in the recovering landscape. Species that avoid one another in the threatened landscape are more likely to respond differently to threat elimination as their resource requirements (and hence threats) probably differ.
After threat management, our prediction was that species that co-occurred with a ‘lost’ species in a threatened landscape were more likely to be lost (due to shared resources or environmental requirements being lost), and that species that co-occur with ‘recovering’ species are more likely to recover (due to shared resources being restored). Using this information, we linked known responses to management by some common well-studied species (eg, the willie wagtail) to unknown responses by rare or hard-to-find species who were able to survive with the known species (eg, the brown treecreeper, a listed threatened species in NSW).
From species to communities
Our study developed the first way of predicting management responses for an entire community of species living together in landscapes where multiple threats act in concert (Tulloch et al, 2018). Our approach enables managers to discover which species will benefit or decline under a selected management strategy, and how the entire community of linked species will change if one or more recovery actions are carried out in a degraded landscape.
We predicted the likely responses to threat management of all 88 bird species persisting in remnant patches of Grassy Box woodland of south-eastern Australia. This endangered ecosystem is being impacted by globally important threats of habitat loss, competitive displacement of woodland birds (by the overabundant noisy miner) and intensive livestock grazing. The innovative approach we developed combines dynamic models of species’ responses to independent and combined threat management actions with network analyses revealing species’ contributions to their co-occurrence network.
Our approach required two sets of information: models predicting the expected change in a species’ colonisation of patches if one or more threats were reduced, and information on how that species shared existing space and resource with other species. ‘Shared space’ was estimated using a method called co-occurrence analysis. This calculates the likelihood that two species will be found in the same patches.
Making more of partial information
Co-occurrence analysis relies on basic information on the presence or absence of each species in every patch, and therefore can be estimated for most species in a landscape. Models of species responses to management, however, generally require surveys repeated over temporal or spatial gradients of change in a level of a threat, and require a certain level of species detection to ensure a good model can be produced that relates the species’ occupancy of the landscape to variation in the threat. This is really hard to achieve for most species, and we were lucky to be able to produce models for 37 of the 88 species detected in more than 1% of surveys. Thus, we used only partial information on the responses of 42% of the bird community to predict change in 100% of the species under alternative management actions of either reduced livestock grazing, improved tree cover or reduction in an overabundant native competitor, the noisy miner.
We were fortunate to have the opportunity to evaluate our predictions five years after management of the three threats to woodland birds; and what we found supported our models. We found that species’ responses to management differed depending on how they were connected with expected ‘increasers’ or ‘decreasers’. If the known species was predicted to increase under management, an ‘unknown’ species that co-occurred with the known species was also more likely to increase. Likewise, if the known species was predicted to decline under management, an ‘unknown’ species that co-occurred with the known species was more likely to decline.
Our paper shows that analyses of co-occurrence networks are crucial for informing decisions about threat management when there are uncertainties about which species might benefit (or suffer) from a given action and not enough time or money to learn about every individual species’ response. By thinking not only about individual species but about how they share space and resources with others, we can ensure that management actions are chosen that benefit the most vulnerable species, and avoid actions that might lead to unintended declines.
More info: Ayesha Tulloch firstname.lastname@example.org
Tulloch AIT, Iadine Chadès & DB Lindenmayer (2018). Species co-occurrence analysis predicts management outcomes for multiple threats. Nature Ecology & Evolution 2: 465-474.https://www.nature.com/articles/s41559-017-0457-3