Introduction

The loss of tree cover has had a greater effect on the functioning of the Earth system than any other change in land surface (Sterling and Ducharne 2008), because woody vegetation provides ecosystem services including carbon storage as well as important habitats for biodiversity. Changes in temperature and precipitation regimes are likely to affect woody vegetation structure, species composition, habitat quality, productivity, and carbon storage (Iverson and Prasad 2002; Broadmeadow et al. 2005; Littell et al. 2010; Good and Caylor 2011; Heubes et al. 2011; Becknell et al. 2012; Fauset et al. 2012; Matías and Jump 2012; Murphy and Bowman 2012). Projections of how woody canopy cover may respond under scenarios of climate change at a landscape scale are poorly resolved, as they are based on gross outputs of dynamic global vegetation models (DGVMs) or extrapolations from historical trends typically discerned from permanent plots or sequences of aerial photography (Fensham and Fairfax 2003; Phillips et al. 2008; Sitch et al. 2008). These approaches suggest disparate trajectories in canopy cover of woody vegetation; for instance, tropical rainforest is vulnerable to cover loss because of reduced precipitation, thermal stress, and increased fire activity; yet, shrub encroachment is evident in many areas (Eldridge et al. 2011), and woody cover in some tropical savannas has increased (Mitchard et al. 2009; Bowman et al. 2010). To effectively predict future climate-related trends in tree cover requires projections from general circulation models (GCMs). For topographically diverse regions with a range of projected climate responses across a small area, such as the island of Tasmania, there is additional benefit from examining climate projections at a finer spatial resolution, generated using a technique known as downscaling (Grose et al. 2010).

Here, we investigate how canopy cover of Tasmanian temperate eucalypt forests is likely to change under a scenario of high greenhouse gas emissions and marked global warming [the A2 scenario from the IPCC Special Report on Emissions Scenarios (SRES) of Nakićenović and Swart (2000)]. We do this by building a predictive model of canopy cover in eucalypt forests in southeastern Australia and then applying this to Tasmania using global climate change data downscaled to a high spatial resolution. This approach is based on the assumption that it is possible to predict with reasonable accuracy the amount of tree canopy cover in natural ecosystems using climate data. While many studies have used such relationships to predict broadscale changes in vegetation distribution and canopy cover in relation to the outputs of GCMs [e.g., in tropical (Hoffmann et al. 2002; Calef et al. 2005) and boreal (Heubes et al. 2011; Tchebakova et al. 2011) regions], there are few regionally detailed projections, reflecting the coarse spatial scale of observed datasets, GCMs, and analysis techniques.

Eucalypts (Eucalyptus, Corymbia, and Angophora species) are unique in being a single lineage that dominates most of the woody vegetation on the Australian continent. Eucalypt forests and woodlands constitute 79 % of Australia’s forest areas and cover 116 million ha (Bureau of Rural Sciences 2008a) in all but the driest parts of the continent. The island of Tasmania is floristically similar to southwestern Victoria, on the Australian mainland (Gill et al. 1985), which is unsurprising, given the two have been periodically connected at times of low sea level during the Quaternary (Lambeck and Chappell 2001). Eucalyptus-dominated vegetation forms a structural continuum of varying height and canopy density in response to variation in climate, soils, and topographic position (Bureau of Rural Sciences 2008b). It is therefore possible to derive a relationship between canopy cover and climate that is common to Tasmania and the southeastern mainland of Australia, even though Tasmania is further from the equator, and cooler and wetter than most of mainland Australia (see Supplementary Material).

Tasmania is a topographically complex island, resulting in a climate with high spatial heterogeneity. Because of the predominance of westerly airstreams, there is a steep rainfall gradient with high year-round precipitation on the mountainous west coast, a drier temperate climate on the less mountainous east coast, and a rain shadow in the Midlands, a north–south valley in the center of the island. Here, in uncleared areas, there are dry eucalypt sclerophyll forest and eucalypt savanna. By contrast, at the wetter end of Tasmania’s climatic spectrum is found the world’s tallest angiosperm, Eucalyptus regnans, in the most carbon-dense forests on earth (Keith et al. 2009).

Tasmania’s future climates are projected to become more like those presently experienced in mainland Australia due to rising temperatures, as well as changes to wind patterns and rainfall caused by a poleward migration of the subtropical high-pressure systems and westerly storm tracks (Cai et al. 2003; Yin 2005). Dynamically downscaled climate projections, with marked regional detail, have recently become available for Tasmania (Corney et al. 2010). This, combined with the existence of similar Eucalyptus-dominated vegetation in warmer and drier climates in mainland Australia (Gill et al. 1985), invites spatially detailed modelling of vegetation cover for Tasmania under various climate change projections.

In this study, we address three specific questions in relation to the effects on eucalypt forests of the changes in climate expected during this century under the A2 scenario and three GCMs:

  1. 1.

    Will there be substantial changes in the canopy cover of Tasmania’s eucalypt forests?

  2. 2.

    Which areas are consistently predicted (by two or three of the three selected GCMs) to increase or decrease in their canopy cover?

  3. 3.

    Are the current eucalypt species likely to tolerate the projected changes in climate, or will translocation and importation of species from mainland Australia be required to maintain eucalypt canopy cover in Tasmania?

Methods

Estimating current canopy cover

In order to generate a climate envelope that accommodates the potential future climate of Tasmania, in constructing our canopy cover–climate model, we used the eucalypt forest domain of the entire southeast Australian region, which includes areas that are hotter and drier (Appendix 1), and with sparser tree cover, than currently exists in Tasmania. Our calibration domain was defined as areas of native vegetation with >20 % tree cover (“forest” sensu Bradshaw, 2012) where the tree layer is dominated by eucalypt species, in New South Wales, Victoria, the Australian Capital Territory, and Tasmania. The integrated vegetation cover dataset (Thackway et al. 2004) was used to remove all areas not representing native forest, such as cropland, grassland and plantations, built-up areas, and water. The National Vegetation Inventory System 3.0 (Executive Steering Committee for Australian Vegetation Information 2003) was used to remove all non-eucalypt vegetation types. Tree canopy cover data (i.e., the percentage of sky light in a plane orthogonal to the ground which is intercepted by trees) at 500-m resolution were then extracted for the remaining grid cells from the MODIS Continuous Vegetation Fields dataset (Hansen et al. 2006). We validated these data against the measurements of eucalypt cover in southern Tasmania (Prior et al. 2013). The final dataset comprised 1,041,030 grid cells representing forest, of which 99,668 were from Tasmania (Fig. 1).

Fig. 1
figure 1

Tree canopy cover for the eucalypt domain in southeastern Australia, derived from the MODIS continuous vegetation field product. White areas indicate non-eucalypt systems and were not considered in our analysis

Full size image

Modelling canopy cover versus climate

Current climate surfaces, representing the mean values of the period 1921–1995 for temperature and rainfall and the period 1970–1995 for evaporation, were generated at 1-km resolution in ANUCLIM version 6.1 (Hutchinson and Xu 2011). Topographic variables were calculated from the NASA SRTM (Rabus et al. 2003) global dataset at a 90-m resolution. The potential explanatory variables used in our modelling are listed in Table 1.

Table 1 List of potential explanatory variables used in modelling of canopy cover
Full size table

Preliminary modelling used generalized linear modelling (GLM), multi-model inference, and model selection using the Akaike information criterion (AIC), which balances model fit with parsimony (Burnham and Anderson 2002) to select a limited, biologically relevant suite of variables to explain canopy cover of eucalypt forest in southeastern Australia: mean annual precipitation, mean maximum January temperature, slope and annual water balance (Mean Annual Precipitation − Annual Pan Evaporation) (Table 1, Appendix 2). Pairwise linear regressions between these variables indicated a strong correlation between mean annual precipitation and annual water balance, but a scatterplot indicated that the relationship was quite nonlinear, so both variables were retained. To develop a canopy cover model based on the present climate, and thereby project future cover under three climate change scenarios, we used Random Forest modelling, which can cope with the potentially strong nonlinear and interacting effects of the selected variables. Random Forest modelling is an ensemble classifier, which generates many classification or regression trees based on bootstrapped samples of the data and aggregates their results (Liaw and Wiener 2002). It is robust against overfitting and has only two parameters (the number of variables in the random subset at each node and the number of trees in the forest), and is usually not very sensitive to their values. By allowing nonlinear responses, Random Forest modelling can also provide a better fit to the data. Random Forest modelling was performed using the “randomForest” package version 4.6-6 in R (Liaw and Wiener 2002) using the four variables from the combined GLM, 500 trees, and one variable per split, on a random subset of 40,000 grid cells to ensure computational efficiency.

Future climate and canopy cover

Future climate projections were based on output from three Conformal Cubic Atmospheric Model (CCAM) dynamically downscaled GCMs (see below) run under the SRES A2 emission scenario. This scenario assumes continuously increasing global population and regionally oriented economic development and represents the highest emission scenario for which a full set of model outputs are available. It is expected that there would be a climate response that is similar in nature but smaller in magnitude under a lower-emission scenario. The three GCMs were (1) CSIRO-Mk3.5, (2) MIROC3.2(medres), and (3) UKMO-HadCM3. These models were chosen for their ability to model current southeast Australian climate means and variability, and to represent the spread of the available model results; this is particularly important when projecting annual precipitation. MIROC3.2(medres) projects an increase in rainfall over southeast Australia greater than the mean of all models considered in the CMIP3 archive, while CSIRO-Mk3.5 projects a greater rainfall decline than most models, and UKMO-HadCM3 shows a moderate decline in mean annual rainfall similar to the mean of all models. The three models together therefore provide a good spread of potential climate projections for the broad southeast Australian region. However, all three models are similar to all CMIP3 models in placing Tasmania near a borderline between increasing rainfall at higher latitudes and decreasing rainfall in southern mainland Australia (Christensen et al. 2007). The exact position of this boundary will determine the specific rainfall change projected for Tasmania, and this is influenced by the downscaling process.

The output of the three GCMs was dynamically downscaled using the regional climate model Conformal Cubic Atmospheric Model (CCAM) of McGregor and Dix (2008) to a final spatial resolution of 0.1 degrees (approximately 10–14 km) grid. Details of the specific modelling methods are found in Corney et al. (2010). Dynamically downscaled regional climate models capture the climate processes that act at the regional scale, rather than assuming a statistical relationship between large-scale patterns of climate elements and local climate, or simply interpolating coarse-scale model output to the local scale. They can potentially capture the regional variation in response to a warming climate, which is especially relevant to Tasmania, given its complex topography, effect of coastlines, and diversity of regional climate influences. Dynamically downscaled models can also simulate extremes more realistically than the GCMs (White et al. 2013). Each of the three downscaled climate projections showed a variety of climate response changes across Tasmania, especially in seasonal rainfall, with the details varying among the GCMs, as described by Grose et al. (2010). For example, MIROC3.2 (medres) projected a rainfall decline in southern Tasmania, UKMO-HadCM3 projected a rainfall increase in most areas, and CSIRO-Mk3.5 projected a mixture of small increases and decreases in a pattern similar to the model average (Grose et al. 2010).

Mean climatic change grids were calculated from each CCAM output for the future period 2070–2099 relative to a baseline period of 1961–1990. Change grids calculated included mean daily T max (°C), mean daily T min (°C), proportional change in rainfall (%), and proportional change in potential evaporation (%) for each month and annual means. The change grids were applied to the ANUCLIM surfaces of mean current climate to calculate future climate surfaces at the same resolution as the input data. Changes were applied as an addition for temperature, and as a multiplication for rainfall and potential evaporation.

The future climate surfaces generated using downscaled climate projections from the three GCMs were then fed into the Random Forest model to generate corresponding projections of future canopy cover for Tasmanian eucalypt forests. Mean canopy cover change was calculated as the change in percent canopy cover in each climatic grid cell. The output from each of the three models was examined to quantify the area of existing eucalypt forest likely to experience a “substantial” increase or decrease in canopy cover. A “substantial” change in canopy cover was defined as a proportional change in canopy cover of >10 %, e.g., a change from 30 to 26 % canopy cover) between the estimated current forest and the three future projections. Thus, canopy cover was defined as “Decreasing” (>10 % decrease), Stable (between −10 and 10 %), or “Increasing” (>10 % increase). We identified areas that were consistently projected to show substantial increase or decrease in canopy cover, which we considered to be those where at least two of the three models agreed in their predictions. Significant deviations from the mean projections for all GCMs were determined using the ImageDiff function of the “SDMTools” package version 1.1-13 in R (VanDerWal et al. 2012), which determines the significance of the pairwise differences relative to the mean and variance of all differences between the two input datasets. The mean projected percentage change in Eucalyptus canopy cover for each Tasmanian Interim Biogeographic Regionalization for Australia region (Environmental Australia 2000) was calculated for each model, as well as the total area subject to substantial projected change under each model across Tasmania.

Species distributions

There are 29 Eucalyptus species in Tasmania, of which 16 are endemic (Wiltshire and Potts 2007). To determine the projected change in climate suitability for Tasmanian eucalypts, and cover the expected range of species responses, we selected one example each of an endemic and a non-endemic species with (1) a widespread distribution in Tasmania; (2) a narrow, high-altitude distribution in Tasmania; and (3) a narrow, low-altitude distribution in Tasmania. For each of these six species, we extracted distribution records from the Atlas of Living Australia (http://www.ala.org.au) and the corresponding January maximum temperature and mean annual rainfall for all locations (including mainland locations) using ANUCLIM version 6.1 (Hutchinson and Xu 2011). These current climate envelopes were then compared with the projected climate envelope for all of Tasmania in 2070–2099, to show the likely change in area with a suitable climate for that species.

Results

Current and future climate envelope

Our analysis shows that Eucalyptus forest currently occupies 33,340 km2 of Tasmania, spanning a range of mean annual precipitation (MAP) from 537 to 2,866 mm and mean maximum January temperature from 15.5 to 23.2 °C (Table 2). Including the data from Victoria and NSW extended the range of MAP down to 257 mm, and mean maximum January temperature up to 36.1 °C (Table 2), thereby including the hottest, driest, and all but the very wettest future climates projected for Tasmanian eucalypt forests in this analysis.

Table 2 Ranges of current climate attributes in Eucalyptus forests in SE Australia (used for the Random Forest model) and Tasmania, and the future projections for Tasmania in 2085 under the three general circulation models (CSIRO Mk3.5, UKMO-Had CM3, MIROC 3.2 Medres)
Full size table

The three CCAM simulations agreed closely in their temperature projections, but differed substantially for precipitation and water balance (Table 2). The simulation using CSIRO-Mk3.5 as input projected only small changes in mean annual precipitation and annual water balance, the simulation using UKMO-HadCM3 projected generally increased rainfall and a wider spread of water availability, while the simulation using MIROC3.2(medres) projected less rainfall and lower water availability over most of southern Tasmania (Table 2). Each of these future climates is seen as an equally plausible climate response given the A2 forcings, but not representative of all possible climate responses.

Current canopy cover

The Random Forest model, which included the terms MAP, January mean maximum temperature, annual water balance, and slope, explained 80.2 % of variance in actual canopy cover. This is a substantial improvement on the generalized linear model fit of 71 % (Appendix 2), reflecting the nonlinear relationships among the variables. Partial dependence plots of variables reveal the existence of thresholds and other nonlinear relationships between the selected environmental variables and canopy cover (Fig. 2). Canopy cover increases steeply between about 500 and 1,000 mm of precipitation and then levels off, and it declines above a mean maximum January temperature of 25 °C. There is an additional very strong decline in canopy cover where the water balance falls below zero (i.e., annual rainfall drops below pan evaporation). The effect of slope is comparatively subtle, with only a slight increase in canopy cover as slope increases from 0° to about 20°, and negligible change with steeper slopes.

Fig. 2
figure 2

Partial dependence of canopy cover (%) on slope, January maximum temperature, mean annual precipitation (MAP), and water balance, based on the Random Forest model. The tick marks on the x-axis indicate deciles of the model training data. The partial dependence function indicates how canopy cover responds to changes in the particular variable after accounting for the average effects of all other variables in the model (Liaw and Wiener 2002). The range shown in these plots indicates the relative strengths of the different predictor variables

Full size image

There was good spatial agreement between the Random Forest model projections and satellite measurements of existing canopy cover for Tasmania (Appendix 3), although with some overprediction around the boundary of the agricultural Midlands region. This can be attributed to the extensive clearing and grazing by introduced animals in these areas, reducing the satellite-detected canopy cover below the climatically expected cover. Although the satellite-derived tree cover data provided only a snapshot of cover, we believe that the effects of fire and forestry have had little effect on the model prediction given the widespread but moderate level of occurrence of both fire and tree clearance across southern Australia, and the large number of sample points used, with non-native vegetation excluded.

Projections of future canopy cover

The three GCMs give a similar projection of the change in overall canopy cover in Tasmanian eucalypt forests for 2070–2099, but showed important differences in the spatial pattern of projected change (Fig. 3). All models suggest that canopy cover should remain fairly stable across most (69–75 %) of its range in Tasmania. However, all three models showed decline in canopy cover in the Eastern Tiers, due to the expected decreases in mean annual precipitation and increases in mean maximum January temperature in this region (Figs. 3, 4). Conversely, all models showed an increase in canopy cover in the central highlands as a result of projected increases in temperature in this coolest part of the state. The models diverge considerably in other regions (Fig. 5): two show substantial tree cover decline in 20–25 % of the current area and an increase in 5–6 %, while the remaining model, CCAM -UKMO-HadCM3, projects the converse: a substantial increase in canopy cover in 22 % of the current area and a decrease in 4 % (Table 2). Specifically, the CCAM-CSIRO-Mk3.5 GCM projections showed the greatest area of decline, concentrated on the southern Midlands, Eastern Tiers, and northeast coast, with some increased canopy cover in the central highlands. The CCAM-MIROC3.2(medres) GCM projection showed a similar area of decline around the Midlands and Eastern Tiers and increase in the central highlands, but stable canopy cover on the northeast coast, while for CCAM-UKMO-HadCM3, the model distinct in projecting increased precipitation on the east coast showed a relatively small area of decline in the Eastern Tiers and increased canopy cover for much of the rest of the Midlands and central Tasmania. Projected mean change in canopy cover for defined Interim Biogeographic Regionalization for Australia 5.1 regions (Environment Australia 2000) regions (Table 3, Fig. S6) shows most consistent decline in the Tasmanian Northern Midlands and Tasmanian southeast regions, although again the higher-rainfall CCAM-UKMO-HadCM3 projects increase in these areas. Canopy cover of areas of tall eucalypt forest is projected to be relatively unaffected by climate.

Fig. 3
figure 3

Present canopy cover in eucalyptus-dominated vegetation in Tasmania (top) and projected cover for the period 2070–2099 under three GCM scenarios (left-hand panels). Areas shown in white are non-eucalypt systems and were not considered in our analysis. The right-hand panels show areas with a substantial change in tree canopy cover for each scenario (defined as a relative change of >10 % of the existing cover, e.g., from 30 to 26 % cover = decrease). The top-right panel shows rainfall isohyets, high elevation areas, and regions mentioned in the text for Tasmania

Full size image
Fig. 4
figure 4

Map showing areas of agreement in projected canopy cover in Tasmania in 2070–2099, based on the projections of the three downscaled GCMs

Full size image
Fig. 5
figure 5

Significant pairwise differences (p < 0.05) in projected eucalyptus cover change for Tasmania under three GCMs, relative to the mean projected change (p < 0.05)

Full size image
Table 3 Projected percentage change in Eucalyptus cover in Tasmanian Interim Biogeographic Regionalization for Australia (IBRA) bioregions under three GCMs
Full size table

Species distributions

There was considerable overlap between the current climate envelope and Tasmania’s future climate for the three non-endemic species (Fig. 6). However, much of the climate envelope of the endemic species falls outside that projected for Tasmania in 2070–2099. Only a few populations of the subalpine species E. vernicosa currently occur in a climate as warm as projected for Tasmania in 2070–2099, and the current, very narrow range of E. morrisbyi is slightly cooler and drier than any area of Tasmania will be in future under the A2 emission scenario (Fig. 6). Even the widespread endemic E. amygdalina is not currently found in areas as warm as most of Tasmania is expected to be in 2070–2099.

Fig. 6
figure 6

a Occurrence in relation to mean annual precipitation and mean January maximum temperature of six eucalypt species, sourced from the species records in the Atlas of Living Australia (http://www.ala.org.au) and ANUCLIM. Tasmanian endemic species are shown in the left-hand panels and non-endemics in the right-hand panels. Tasmanian records are represented by squares and mainland records by crosses. The gray area represents the climate envelope projected for Tasmania for the period 2070–2099 under the CSIRO Mk3.5 Model. Eucalyptus amygdalina and E. obliqua are widespread, while the other species have more restricted distributions in Tasmania. b Climate envelope for the CSIRO Mk3.5 projections compared to the UKMO-HadCM3 and MIROC3.2 (medres) model projections

Full size image

Discussion

Projected changes in canopy cover in Tasmania’s eucalypt forest

Our study shows that the climate across most of Tasmania will continue to be suitable for eucalypts over the next seven or more decades under the A2 emissions scenario with marked climate changes, but that climate change is likely to alter the distributional patterns of tree canopy cover in Tasmania’s eucalypt forests. These changes are influenced by the island’s complex topography and climate, and their details vary according to the GCM used. Our analysis is admittedly broad-bush, and it is possible that shifts in climate variables not included in our model, for instance, changes to seasonal patterns of extreme temperatures and precipitation, might alter eucalypt tree cover in specific areas. Such complexities are difficult to capture in prognostic analyses using projected climates; hence, we chose to use a suite of climate variables selected because they were biologically meaningfully and readily available from climate models outputs. However, we believe the inclusion of the July and January monthly temperature extreme values helped to take into account this putative effect (Appendix 4).

Two of the three GCMs we used suggest that conditions will become hotter and, on average, drier, leading to a net loss of canopy cover in the state’s eucalypt forests. All GCMs indicate this canopy cover loss will be centered on the Midlands, the driest part of the state. There is also a smaller area, focused on the Central Plateau, where canopy cover may potentially increase as temperatures warm from the current sub-optimal ones.

The analysis indicates that the low-lying regions of the east and northeast have the largest potential for decline in eucalypt cover. This region is in the lower range of mean annual rainfall for Tasmania (generally 600–1,000 mm), with mean annual potential evaporation at ~1,000 mm, creating a mean water balance of approximately 0 to −400 mm. The tree cover model shows eucalypt cover has high sensitivity to rainfall in the range 500–1,000 mm and water balance between 0 and −500 mm, so the projected change in tree cover in this region is driven mainly by the projected change in mean annual rainfall and water availability. There is a spread of projections in rainfall for this region in the three models examined, with two showing a decline and one showing an increase, and also an increase in rainfall indicated in this region by the other three CCAM simulations not considered (Grose et al. 2010). There is also a high degree of uncertainty when examining projected rainfall changes at this highly regional scale. Therefore, rather than giving a single central projection of eucalypt cover in this region under a warming climate, we highlight the sensitivity of cover to water balance in this region and indicate that there is a range of possible projections, including a plausible scenario where eucalypt cover declines by >10 % based on the changes to mean conditions alone. GCM selection is therefore crucial, and interpretation of projections based on any individual GCM should acknowledge this source of uncertainty. However, there are other factors to consider not covered by changes to the mean.

Lower tree canopy cover and biomass may result from one or a combination of lower recruitment rates, increased mortality and slower growth. Tree recruitment is likely to be particularly vulnerable to the direct and indirect effects of climate change because woody seedlings are sensitive to drought, climate variability, fire, and herbivory (Brown and Wu 2005; North et al. 2005; Ibanez et al. 2007; Lloret et al. 2009; Brodie et al. 2012; Kouba et al. 2012; Mok et al. 2012). Globally, there is evidence that tree mortality is increasing because of more frequent and more severe droughts and fires (van Mantgem et al. 2009; Allen et al. 2010; Michaelian et al. 2011; Williams et al. 2010), leading to reduced woody vegetation cover in many regions, especially semiarid ones (Notaro 2008). In addition to changes in average rainfall, increasing climate variability and higher evaporation rates from hotter temperatures will also reduce water availability to plants. Severe drought, which is likely to become more frequent with climate change, has been linked to substantial crown dieback and death of eucalypts in mainland Australia (Fensham et al. 2009; Brouwers et al. 2012). Drought also exacerbates the dieback and mortality of Tasmanian eucalypts induced by intensified land-use practices (McMurray 1983; Davidson et al. 2007).

The observed trends in eucalypt canopy cover in relation to climate are similar to global patterns showing a peak in forest biomass in cool maritime climates (Larjavaara and Muller-Landau 2012). The climatic trends in canopy cover were also very similar to those of eucalypt growth (Prior et al. 2011), consistent with growth rate being a major determinant of canopy cover. In warm climates, tree growth is limited more strongly by low precipitation and high temperatures than by low temperatures (Mäkinen et al. 2002; Reich and Oleksyn 2008; Maxime and Hendrik 2011). Thus, as temperatures increase, growth of eucalypts is likely to slow in all but the coldest areas of Australia (Prior et al. 2011). It is unlikely that these climatically induced growth reductions will be offset by the CO2 fertilization effect, even though water-use efficiency will probably increase (Peñuelas et al. 2011; Booth 2013).

Indeed, over the past 50 years, increased atmospheric CO2 has consistently led to increased water-use efficiency across the planet but tree growth has generally declined, especially at lower latitudes (Silva and Anand 2013). Any direct benefits of increased atmospheric CO2 on tree growth may have been offset by perturbations to foliar nutrient concentrations (Sherwin et al. 2013), as well as the climatically induced growth reductions discussed above.

Our modelling considered changes in mean annual precipitation, but not in the seasonal or inter-annual distribution of that rainfall. The changes in seasonal rainfall are expected to be larger than annual totals, with the west coast showing an increase in winter and a decrease in summer rainfall, the central highlands a decrease in every season, and the northern east coast an increase in autumn and summer rainfall (Grose et al. 2010). Trees growing in areas with decreased summer rainfall will experience more intense drought stress, potentially leading to greater loss of canopy cover than suggested by our analysis. In addition, increased inter-annual climate variability may reduce vegetation cover (Notaro 2008).

On continental or global scales, while climate constrains the upper limit of forest biomass, it explains little of its variation, with realized biomass often falling below the constraint due to disturbances such as fire and herbivory (Sankaran et al. 2008; Stegen et al. 2011). The effects of fire are implicitly included in our analysis, because current tree canopy cover is modified by fire. This is based on the assumption that climate and fire regimes are in a dynamic equilibrium. However, any changes to the current fire–canopy cover relationship (for example, due to changes in rainfall seasonality) may alter future canopy cover from the values projected here. In particular, tall eucalypt species are especially vulnerable to increased frequency and intensity of wildfire because most are obligate seeders and generally killed after fire (Tng et al. 2012).

Will species translocations be required? Responses of individual species

Our projections of canopy cover change are probably optimistic, because they are based on cover of species and provenances that are adapted to current, rather than future, climates and assume instant dispersal and establishment of trees to new regions. Our analysis effectively assumes that many existing tree species are replaced by ones better adapted to the warmer conditions. Hughes et al. (1996) stressed the fine thermal tolerance of most eucalypt species, with half of all species having a range in mean annual temperature across their distribution of less than 3 °C. Tasmania’s future climates are already experienced by eucalypts on mainland Australia (Fig. 4), but may fall outside the current climate envelope of some Tasmanian endemics, which are therefore likely to be most affected by climate change. Increasing drought frequency has already been implicated in the decline of the endemic Tasmanian tree Eucalyptus gunnii spp. divaricata (Sanger et al. 2011), and other endemic species are also likely to be adversely affected by the hotter, drier summers projected for much of the state. Therefore, to achieve the future canopy cover projected by our models, compensatory range expansions by other species are required. This necessarily involves substantial lags unless anticipatory planting of non-native provenances or species is undertaken (Broadmeadow et al. 2005). Currently, there are plant trials and genetic studies to identify Tasmanian eucalypt species suitable for restoration plantings (Bailey et al. 2013). Restoration and carbon sequestration plantings need to consider the species and provenances most suitable for future, rather than present, climates (Close and Davidson 2004).

Not only do tree species vary in their climatic requirements, but also in their responses to changing climate (Linares et al. 2011). Already, increased aridity and fire frequency in eucalypt forests in southwest Western Australia have shifted species composition toward fire-resistant species and resulted in denser stands of small trees (Pekin et al. 2009). Similar changes in species composition and structure can be expected in Tasmanian forests, with concomitant changes in tree canopy cover and carbon storage.

To conclude, while eucalypt forests should persist over much of Tasmania during the coming century, climate change will modify their distribution, structure, canopy cover and biomass, and species composition. Maintaining the current high levels of canopy cover may require planting of new Eucalyptus species and provenances, including those currently found only on mainland Australia.