As the climate changes, it may be possible to maintain areas of lowland grasslands by applying alternative management options to related communities, such as the Bursaria-Acacia woodland and scrub community (NBA) and the Lowland Grassland Complex (GCL). NBA is part of the natural continuum from grassland to woodlands, so the application of a regular, low-intensity fire regime to this community could promote grassland vegetation and reduce shrub cover. This may be appropriate within all grasslands if increased woody growth occurs, if a primary conservation objective is to maintain grassland vegetation structure. The derived grasslands, which are the result of clearing native grassy, dry woodland and forest communities, could be managed to improve movement across the landscape. Given the highly fragmented nature of the landscape, these areas may be essential to maintain connectivity for future movement of genes and species. However, these options may only be successful in maintaining depauperate examples of lowland grassland, and may not conserve the floristically and genetically diverse grasslands of greatest conservation significance.

In areas outside the current distribution that are projected to become climatically suitable in the future, the management response could include active translocation or restoration of grassland species, to increase landscape connectivity and conserve species of particular social or cultural value. However, the emphasis should be directed towards species or ecotypes expected to be more tolerant of new conditions [ 62]. In locations where the climate becomes unsuitable for the dominant species, weed control will be important to support the transition to new native dominant species.

Managing for healthy ecosystem function, rather than primarily for species composition, fits well within a risk management framework because there will always be uncertainty associated with projections of future climate and in predictions of how animals and plants will respond to a changing climate [ 49, 63]. Firstly there is the uncertainty associated with the range in climate models, each of which is considered to represent a plausible future [ 64]. The six models used here all perform well based on metrics of rainfall compared with observations in south-east Australia [ 35], yet they project different extents and locations of climatically suitable areas. The projections for the GTL community, for example, ranged from no suitable climate remaining, through to small areas remaining in different localities. These differences highlight the importance of considering results from a range of climate models to represent a range of plausible future climates and trajectories for a community or species.

However, making decisions on the basis of multiple distribution maps poses a challenge for conservation planning. The method used to combine the maps (e.g. taking the mean, the median, the bounding box or the minimum overlap of the different models) will influence interpretation of the results. This can have consequences where conservation effort is deployed across the landscape since both extent and location of projected suitable climate will differ between climate models. Over-confidence in a single map risks the possibility that conservation investments are applied to areas that do not support the target community or species under future climate. A precautionary approach to incorporate the uncertainty due to the range in climate models is to prioritise conservation in areas where model agreement is high, while acknowledging that all climate models project plausible futures [ 65]. In contrast, the bounding box, the sum of all areas projected to be suitable by all models, is likely to give an overestimate of the extent of future climate suitability, since each single model projects a substantially smaller area. Alternatively, considering the ‘best’ and ‘worst’ case scenarios may be a useful framework for decision making.

Future rates of greenhouse gas and aerosol emissions are another major source of uncertainty in projections of future climate suitability. We present results based on the high SRES emissions scenario (A2), which is a more likely scenario than lower emissions scenarios for the coming decades given current trajectories of global emissions [ 66]. We are currently tracking at the higher end of the A2 scenario [ 39] and even if drastic mitigation measures were achieved now, we will be dealing with the ‘committed warming’ that is now built in to the Earth’s climate system. However, while current emissions restrict the likelihood of the different scenarios in the short- to medium term, it does not necessarily mean that the scenario will be followed to the end of the century [ 67]. If emissions were mitigated the changes to the grassland communities for 2080 could be expected to be similar in nature but smaller in magnitude than those presented here for the high emissions scenario.

Finally, there is uncertainty associated with the choice and parameterisation of the ecological model used. Variable choice in species distribution models can affect the projections of current and future suitability [ 28, 30, 31]. Using a saturated SDM has the potential to underestimate suitable area, although Maxent has been shown to be relatively robust to correlated variables [ 32], and attempts to avoid over-fitting by applying a regularization penalty for each term included in the model [ 22, 33, 34]. The inclusion of non-climate predictors may provide more realistic distributions when knowledge of the biology is available. However, climate-only SDMs have been shown to be effective and efficient for initial assessments of climate suitability compared to models based on climate and additional predictor variables [ 68]. We chose a climate-only approach because our objective here was to highlight the potential for changes in the climate domain of the community to occur under future conditions. If the objective were to prioritise particular areas for management, soil, slope, aspect and other biophysical data would be needed to identify suitable habitat within the climatically suitable areas.

Limitations in the available tools for understanding potential change and our inability to predict the future need to be acknowledged when making conservation management decisions [ 69]. In the absence of objective tests of model accuracy [ 28], it is important that the limitations of SDMs in general are understood, and the influence of particular modelling decisions are acknowledged and incorporated into conservation decision making. There are many factors that are known to be important determinants of species distributions that SDMs cannot consider, including dispersal, potential adaptation, species interactions, and potential responses to elevated carbon dioxide [ 70]. With the cascading uncertainties inherent in predicting responses to future conditions, SDMs are just one tool among many that can inform decision making by highlighting the potential for change to occur [ 71].

Current approaches to conservation covenants and biodiversity offsets were not designed to cope with uncertainty, or with changing boundaries or composition of community types. Mechanisms for protecting grasslands currently include formal reservations with fixed boundaries on title. Long-term conservation covenants have also become important conservation policy instruments, whereby landowners forgo environmentally detrimental land-use rights, often in exchange for financial support, while retaining ownership of the land [ 72]. These reserves and covenants are legally required to be managed under a certain regime, with management success judged by indicators such as the abundance of listed species and floristic composition. Our results suggest this may no longer be appropriate under climate change. Legal and policy instruments therefore need to be more flexible to allow for changes in external factors that are not under management control, such as climate change [ 73]. Flexibility could potentially be enabled through supporting regulations. However, safeguards will be required to ensure that increased flexibility does not weaken the protection of species rich communities of high conservation value. Management approaches that are outcomes-based rather than prescriptive may be helpful and allow for innovation. For example, stewardship payments to farmers in return for conserving biodiversity on their farms may be more flexible, and ultimately successful, than proscriptive, regulatory approaches (for example, see the Midlands Conservation Fund ( http://www.bushheritage.org.au/places-we-protect/state_tasmania/tasmanian-midlands)).

A more flexible regulatory and policy environment could facilitate the protection and restoration of grasslands by allowing their transition into transformed or even novel ecosystems. If management objectives include retention of grassy ecosystems, then flexibility about their composition will be required. For example, the current listing under Australian national environmental legislation sets thresholds not only for the degree of weediness and tree cover, but also floristic composition. By the end of the century there may be few grasslands matching the description of the ecological community listed in 2009, but there may well be healthy native grassy vegetation, perhaps with no current analogue, that could be considered a conservation priority.

Our results highlight some fundamental questions about the conservation of threatened communities over the next few decades. While we know that substantial changes are likely, we do not know what will replace these communities as their climate envelope becomes unsuitable. We do not know whether the changes will be gradual and incremental or occur as a sudden threshold shift. We do not know exactly where grasslands of high conservation significance will remain. However, we do know that changes in species composition will occur, and the distribution and composition of ecological communities will shift. This suggests that the current focus on single-species conservation and maintenance of composition should be revised. Prioritising locations of high conservation significance, rather than species or communities defined by their species composition, might better conserve ecological processes and function into the future. The concept of the ecological community will remain useful in conservation if it is seen as a dynamic, rather than a static entity, in which ecological processes and functions are sustained and contribute to patterns of diversity across the landscape.