East Antarctica has shown little evidence of warming to date1,2,3 with no coherent picture of how climate change is affecting vegetation4,5,6. In stark contrast, the Antarctic Peninsula experienced some of the most rapid warming on the planet at the end of the last century2,3,7,8 causing changes to the growth and distribution of plants9,10,11. Here, we show that vegetation in the Windmill Islands, East Antarctica is changing rapidly in response to a drying climate. This drying trend is evident across the region, as demonstrated by changes in isotopic signatures measured along moss shoots12,13, moss community composition and declining health, as well as long-term observations of lake salinity14 and weather. The regional drying is possibly due to the more positive Southern Annular Mode in recent decades. The more positive Southern Annular Mode is a consequence of Antarctic ozone depletion and increased greenhouse gases, and causes strong westerly winds to circulate closer to the continent, maintaining colder temperatures in East Antarctica despite the increasing global average15,16,17,18. Colder summers in this region probably result in reduced snow melt and increased aridity. We demonstrate that rapid vegetation change is occurring in East Antarctica and that its mosses provide potentially important proxies for monitoring coastal climate change.


Climate change is causing many species to shift poleward in response to increasing global temperatures19. However, Antarctic continental vegetation is unlikely to exhibit such simple and predictable responses, as warming is inconsistent over the continent1,3, and species distributions are largely determined by local availability of ice-free habitats and water, rather than temperature per se6,20,21,22. Small changes in the microclimate (temperature, precipitation, wind or humidity) can impact the water balance or freeze–thaw cycles23 and thus impact vegetation, even in the absence of regional warming.

Continental Antarctic flora is mainly restricted to ice-free coastal regions and dominated by lichens and mosses, with two vascular plant species found only on the comparatively mild Antarctic Peninsula20. We have been monitoring health and biodiversity in well-developed moss communities in East Antarctica (Fig. 1; ref. 24). Such moss beds occur only where summer ice melt produces temporary streams and lakes, sustaining a short (8–16 week) season of plant growth4,24,25. While lichens are more abundant across Antarctica, they grow extremely slowly, making them unsuitable for such monitoring.

Fig. 1: Changes in East Antarctic moss communities.
Fig. 1

Periodic flooding of moss beds (1982–2000) favoured S. antarctici and B. pseudotriquetrum, which tolerate submergence (Supplementary Table 1; ref 24); however, C. purpureus, which does not survive extended submergence, was restricted to higher elevations. ASPA 135 (photographed in 1999) illustrates inundation; since then, the moss beds have dried and experience less submergence. C. purpureus can now survive and is more abundant, along with increased moribund moss (see also Fig. 2). Acute stress has caused moss colour change (2008) due to more photoprotective (red–brown) pigments, with some recent recovery (2013; see also Fig. 3).

Over 13 years (2000–2013), we observed significant changes in species composition within moist moss communities in East Antarctica (Fig. 1). Changes in the relative abundance of moss species were modelled from microsamples collected from replicate quadrats at sites on two peninsulas in the Windmill Islands (Antarctic Specially Protected Area (ASPA) 135 (site A2) and Robinson Ridge (site RR)). This sampling captures two of the four major moss beds, which occur on three peninsulas in this region (Supplementary Fig. 1). We found that moss communities were dominated by the Antarctic endemic species Schistidium antarctici (Figs. 1 and 2)—the most submergence tolerant and least desiccation tolerant of the three co-occurring species (see Supplementary Table 1; ref. 21). The relative abundance of S. antarctici was significantly lower than the baseline (2000) in all years between 2008 and 2013 (the mean abundance decline from 2000 to 2013 for ASPA 135 was 93.8 to 63.0%, while for Robinson Ridge, it was 83.4 to 55.6%; Fig. 2). Conversely, the abundance of the two desiccation-tolerant, cosmopolitan species—Ceratodon purpureus and Bryum pseudotriquetrum—was generally higher than the 2000 baseline over the same period (Fig. 2 and Supplementary Table 1). In particular, C. purpureus was significantly more abundant in 2011–2013 compared with 2000 at both sites, indicating drier growth conditions24 (the mean abundance increase from 2000 to 2013 for ASPA 135 was 0.7 to 11.4%, while for Robinson Ridge it was 0.9 to 7.7%; Fig. 2). Given this species’ aversion to submergence21,23,24, it is likely that these sites are now experiencing reduced periods of water inundation, allowing C. purpureus propagules to establish (Fig. 1).

Fig. 2: Community composition at two sites in the Windmill Islands, East Antarctica, sampled over six summer seasons between 2000 and 2013.
Fig. 2

a,b, Results are presented for ASPA 135 (A2) (a) and Robinson Ridge (RR) (b). The moss community consists of varying proportions of three species and unidentifiable moribund moss (lacking cell contents and essentially dead). Relative abundances (±95% CI) were estimated using a Bayesian model based on 9 microsamples per quadrat (n = 10 quadrats in all years except 2000, when n = 7; see Methods). The model also estimated the differences in relative abundances from the 2000 baseline year, with asterisks marking significant changes (*P < 0.05; **P < 0.01; ***P < 0.001).

This apparent drying was further supported by our assessment of changes in vegetation health over the same period. We used digital image analysis to estimate the percentage cover of moss based on the leaf colour (healthy, green; stressed, red–brown; and moribund, grey–black (indicating total loss of photosynthetic pigments)). In 2003, all quadrats contained at least 70% healthy, green moss (Fig. 3), but by 2008, more than half of this healthy moss had turned red–brown in colour (due to drought and high light or other stressors). Red–brown colouration indicates a shift away from photosynthesis and growth (high chlorophyll, green moss) towards investment in photoprotective pigments in response to physiological stress26. Mosses can recover if conditions improve; new healthy moss plants can re-sprout through moribund turf and stressed moss will re-green if photoprotective pigments decline relative to chlorophyll. There was some recovery to healthy moss between 2008 and 2013 (Fig. 3), but recovery differed between the two sites. At Robinson Ridge, stressed moss increased from 5 to 50% between 2003 and 2008, with 30% of the moss remaining in a stressed state in 2013, whereas at ASPA 135, the 2013 percentage covers of healthy (85%) and stressed (3%) moss were similar to 2003 (Fig. 3). However, it is notable that both the microsamples (Fig. 2) and photographs (Fig. 3) suggest that moribund moss increased during the study period—significantly so at ASPA 135. The increased dominance of the two more desiccation-tolerant moss species (C. purpureus and B. pseudotriquetrum) at the expense of endemic S. antarctici (Fig. 2) and the reduction in moss health (Fig. 3) are all consistent with drier microclimates. Given that it can take weeks for the health or composition of mosses to respond to changing conditions in the field21, these changes probably represent differences in seasonal growth conditions rather than specific weather conditions during sampling. While there are variations in microtopography within quadrats due to factors such as frost heave, with stressed and moribund moss more likely to occur on the drier and exposed ridges, there has been no consistent change in the microtopography that would explain a directional shift. The changes in moss health and species composition probably reflect moist microhabitats contracting to lower elevations, both within quadrats and across sites.

Fig. 3: Change in moss health between 2003 and 2013 at two East Antarctic sites.
Fig. 3

The photographs (top) demonstrate how moss health changed in a typical quadrat over time. The graphs show the percentage cover of three moss health categories (healthy, green; stressed, red–brown; and moribund, grey–black) at ASPA 135 (A2, bottom left) and Robinson Ridge (RR, bottom right). Bars represent the mean percentage cover ± 95% CI (n = 10 quadrats per site). Letters denote significant differences (P < 0.05) within health categories. Note that moss cover categories do not sum to 100% as quadrats contain small proportions of rock, snow, ice, water and lichens (data not shown), which did not change significantly over time at either site.

The trend towards drier conditions in the Windmill Islands is also evident in stable carbon isotope (δ13C) signatures along slow-growing moss shoots of all three species collected from six sample locations across the three peninsulas, including both quadrat sites (A2 and RR; Fig. 4 and Supplementary Fig. 1). Antarctic mosses have very slow growth rates and their cellular δ13C signatures have been shown to accurately reflect bioavailable water13. Shoot segments with higher (less negative) δ13C reveal periods when moss was growing under wetter conditions13 (more submergence; see Fig. 1). Changes in δ13C along moss shoots, which can be precisely dated using ‘bomb-peak’ radiocarbon dating10,12 (see Methods), enable interpretation of highly resolved temporal trends in water availability in Antarctica.

Fig. 4: Rate of change in moisture availability in East Antarctic mosses since 1960.
Fig. 4

Rates were estimated using isotopic analysis of 18 long moss shoot cores from 6 sites. Data represent linear trends in δ13C (±95% CI). Negative δ13C change rates indicate that water availability has declined, with significant differences indicated by asterisks (*P < 0.05; **P < 0.01; ***P < 0.001). Squares indicate slow-growing cores with fewer than ten samples available since 1960. Triangles indicate fast-growing cores that span fewer than 20 years. Quadratic responses were also fitted, but only the three cores marked with blue curves were significantly nonlinear.

Windmill Islands moss beds support live shoots, up to 135 mm long, that have been growing for decades12,13 at an average rate of 1.4 mm yr−1 (s.d.: 0.8; Supplementary Fig. 2). Trends in δ13C in moss shoots also indicate that bioavailable water has declined in the Windmill Islands region since the 1960s, independent of species or site. On average, δ13C decreased at a rate of 0.039‰ yr−1 (s.d.: 0.068, n = 18), and this drying trend was significant in 40% of these slow-growing shoots (Fig. 4). Variation in δ13C trends between shoots suggests localized topography results in some moss shoots drying faster than others, with mosses likely to persist longer in moister microrefugia20. That an overall negative δ13C trend was found provides further support for a drying trend in this biologically important polar region.

This regional drying is possibly due to reductions in stratospheric ozone and increases in greenhouse gases causing an intensification in the positive phase of the Southern Annular Mode (SAM; Fig. 5c)15,16,17,18,27. Ozone depletion since the 1970s has had pronounced effects on Southern Hemisphere climate, most obviously over the spring and summer growing season (refs 3,16,28 and the references therein). This has led to increased wind speeds around Antarctica and lower maximum temperatures across much of East Antarctica1,3,16,17. Lower temperatures are associated with reduced snow melt, meaning more water remains frozen and biologically unavailable20.

Fig. 5: Summer (December to February) SAM and meteorological observations for the Windmill Islands since 1961.
Fig. 5

ad, Trends in time series (degree days (a), mean wind speed (b) and SAM index (c)) and the probability of change points (d) were estimated by a Bayesian model (see Methods). Solid lines in ac are the mean, dashed lines represent the 95% CI. Degree days is an indicator of snow melt, calculated as the sum of the maximum temperatures for days when temperatures exceeded 0 °C. e,f, Linear regressions showing significant relationships between the SAM and degree days at all three data collection stations (e) but not between the SAM and mean wind speed (f). Solid and dashed lines in e represent significant trends (*P < 0.05; **P < 0.01, ***P < 0.001), whereas dotted lines in f represent trends that were not significant (NS). Wilkes station is dashed in e because the different slope might indicate an effect of station move.

Ozone depletion and the increase in the SAM3,15,17,18 have been linked in recent years to a growing number of biological phenomena across the Southern Hemisphere, including changing growth rates in trees16,29 and increased body weight and breeding success in wandering albatross30. It seems likely that the Windmill Islands represent another example where regional drying, linked to anthropogenic climate change, is associated with declining plant health and changes to species composition in terrestrial communities, as well as shifts from freshwater species (green algae and cyanobacteria) to diatoms, as nearby lakes become more saline14 (see Supplementary Fig. 1 for locations).

To confirm recent changes in regional climate we compiled long-term weather data from the three Australian Government Bureau of Meteorology (BOM) stations (Wilkes, Casey Tunnel and Casey; Supplementary Fig. 1) that have been operational since 1961. Despite interannual variation typical of polar climates, these records are consistent with regional drying in response to the increasing SAM. The number of degree days above freezing per year (a potential indicator of snow melt) decreased by 31% from 208 (s.d.: 42; 1975–1993) to 143 (s.d.: 44; 1994–2017), largely due to a distinct downward shift in 1993/1994 (Fig. 5a,d). Similarly, there has been a trend towards higher mean wind speeds since 1978 (16.3 km h−1 for 1978–1987 to 21.4 km h−1 for 2000–2017; Fig. 5b,d). The SAM exhibited a steady increase since the 1960s (Fig. 5c), but probable change points in the climate series for the SAM, degree days and wind over the 15 years of moss observations (Fig. 5d) suggest that linear trends do not adequately explain recent data. A more positive SAM was significantly correlated with lower degree days, in particular since 1970, when the correlations were consistent for data from the two consecutive Casey weather stations (Fig. 5e; coefficient of determination, R2 = 0.66, P < 0.001 for the Casey Tunnel and R2 = 0.38, P < 0.001 for Casey). Although we found no correlation between SAM and mean wind speeds (Fig. 5f), there is published evidence for such a link16,17. While lower snow melt and higher winds are plausible explanations for drier conditions in moss beds, actual water availability will be determined by a complex balance of snowfall, redistribution by wind, and subsequent melt and refreezing. Unfortunately, precipitation observations are unreliable given the strong winds and blown snow. Therefore, while BOM data support the drying trend, the evidence is not as strong as the changes in community composition and isotopes, which provide a direct measure of changes in the overall water balance.

In addition to the long-term drying trend apparent in these communities, our results show that a sudden decline in plant health occurred in 2008 (Figs. 1 and 3), followed by partial recovery. This could indicate that an extreme weather event occurred in that season. BOM synoptic records show lower maximum temperatures and degree days in 2008 (Fig. 5a), but there was also an unusual occurrence of freezing rain, which happens when supercooled raindrops fall in liquid phase but freeze when impacting the surface. Over the 25 years from 1989 to 2014, there were only 16 observations of freezing rain. Twelve of those occurred between 2006 and 2009 with three in December 2007 shortly before the quadrats were monitored in January 2008. Although this evidence is circumstantial, it is plausible that an unusual occurrence of freezing rain caused additional stress to the mosses, and given that disturbance can promote community change31, this extreme event may have facilitated the establishment or increase of C. purpureus under this drying regional climate.

The presence of lichen encroachment over moribund moss has previously been considered as evidence for drying conditions in the Windmill Islands; however, the timeframe for this change was unclear24,25. It could represent a slow process that has been occurring over the past 2,000–9,000 years as deglaciation and isostatic uplift have gradually resulted in drier sites24; however, evidence is now accumulating that a more contemporary drying trend may also be impacting the region. Rapid increases in salinity from three Windmill Islands lakes suggest drying across the region in recent decades14, and a time series of photos from Mossel Lake ~1,400 km away in the Vestfold Hills also provides circumstantial evidence that the drying could be more widespread (Supplementary Fig. 3). Together, the combination of lake salinity records, shoot stable isotope analysis, meteorological data and vegetation monitoring paint a clear picture of recent drying across this region leading to community change.

Our findings suggest that a drying trend is affecting East Antarctic terrestrial biota. Antarctic mosses are small and grow very slowly, but their tissues maintain a record of the environmental conditions during growth10. The suite of methods applied here demonstrates that Antarctic moss communities are far more responsive to their microclimate than we might predict based on their slow growth rates; indeed, we show that rapid changes can and do occur in these terrestrial Antarctic vegetation communities.

The ice-free habitats of the Windmill Islands are among the most extensively vegetated areas on the Antarctic continent, with several large moss beds and even larger areas dominated by lichens24,25, rendering the area critical for Antarctic biodiversity. Our data suggest that climate change and ozone depletion are already impacting old-growth moss beds in East Antarctica with other, as yet unmeasured, components of Antarctic biodiversity potentially also affected31. Continued monitoring of these study areas is important to fully understand the impacts on all terrestrial biodiversity and develop appropriate conservation strategies. This is particularly true for the Windmill Islands, as it is the only region of East Antarctica predicted to have an increased number of ice-free areas by 210022, based on a modelled increase in degree days (50–100) equivalent to the decline in degree days reported here for the past 23 years. Furthermore, given the spatial and temporal paucity of meteorological records for Antarctica, moss communities are important and sensitive biological proxies for tracking coastal climate change around the continent.


We monitored vegetation change from 2003–2013 within ten permanently marked quadrats at two sites in the Windmill Islands region, East Antarctica, and combined this with an earlier survey in 2000 of seven quadrats in the same locations24 (as detailed in Supplementary Fig. 1). Moss communities were targeted as the most sensible way to detect change in Antarctic continental vegetation. While mosses are the dominant plants in East Antarctica, lichens are the major component of vegetation at the study site. Lichens cope better with drier conditions but grow much more slowly (0.01–0.07 mm yr1)20. As a result, lichens would be less likely to demonstrate any measureable impacts from drying within 10 years, and arguably by the time such changes were observed the moister moss communities could be irreparably damaged. We estimated the health status of the moss turf from digital photographs of each quadrat. The rate of change in moisture availability since 1960 was also measured using radiocarbon dating and δ13C analysis of 18 long intact moss shoot cores.

Study area and species

This study was conducted in the Windmill Islands of East Antarctica at six sites across three peninsulas (Supplementary Fig. 1). Three sites were on Bailey Peninsula: two in ASPA 135 (A2, 66.283° S, 110.533° E32; and A3, 66.283° S, 110.541° E) and the third adjacent to the accommodation building at Casey Station, Red Shed (RS, 66.283° S, 110.528° E33). Two sites were in ASPA 136 on Clark Peninsula12: Clark 1 (C1, 66.253° S, 110.555° E) and Clark 2 (C2, 66.246° S, 110.592° E). The sixth site was at Robinson Ridge (RR, 66.368° S, 110.587° E33). These sites have all been described previously12,32,33 except for the A3 site, which is located along an ephemeral stream within ASPA 135. The A2 (map ID 14450; http://data.aad.gov.au/aadc/mapcat/display_map.cfm?map_id=14450) and RR (map ID 14451; http://data.aad.gov.au/aadc/mapcat/display_map.cfm?map_id=14451) sites are locations for permanent quadrats used for long-term monitoring of terrestrial vegetation dynamics in the Windmill Islands34,35. Summer melt water in streams and lakes sustains some of the most extensive moss beds of continental Antarctica (for example, Red Shed, ASPA 135, Robinson Ridge and Clark 2 sites36), comprising three moss species: the dominant endemic moss S. antarctici and two cosmopolitan species, B. pseudotriquetrum and C. purpureus. Ancient penguin guano deposits at these sites have contributed to a high density of vegetation24,37.

Small, scattered populations of moss also exist in the more elevated areas; for example, Clark 1. Tolerance to desiccation and submergence varies between these three moss species (Supplementary Table 1). S. antarctici typically grows in wetter sites and is least desiccation tolerant, whereas C. purpureus is not tolerant of submergence and is generally found in higher, drier areas, consistent with its high tolerance of desiccation (Fig. 1)21. In contrast, B. pseudotriquetrum is a relatively plastic species that is found in both dry and waterlogged sites21,38. B. pseudotriquetrum tends to grow and respond faster under favourable conditions13,26,39, whereas C. purpureus and S. antarctici have similar slower growth and response times. Research and sample collection was conducted under the Antarctic Treaty (Environment Protection) Act 1980. Permit numbers ATEP2-12-13-4046, ATEP10-11-12-3042-3129, ATEP07-08-1313, ATEP05-06-2542 ATEP02-03 and ATEP99/10 were issued by the Commonwealth of Australia Department of the Environment to S.A.R.

Sample collection and analysis

Quadrats (25 cm × 25 cm) were located close to ephemeral summer water sources (Fig. 1), such as melt streams and lakes, which sustain verdant bryophyte communities, and positioned where bryophyte turf was dominant. Permanent quadrats were established in 2003 at the same locations as a pilot conducted in 200024. These quadrats continue to be in some of the wettest locations in the region and to support the healthiest moss communities. The presence or absence of moss species was analysed from nine evenly distributed tweezer-pinch samples (referred to in the main text as microsamples) taken from seven to ten quadrats at each site in 2000, 2003, 2008, 2011, 2012 and 2013 (details in Supplementary Table 2). We then scored the presence of live and moribund moss within vegetation samples and identified live bryophytes to species level24.

Long intact moss shoot cores (n = 18; 30–135 mm) of B. pseudotriquetrum, C. purpureus and S. antarctici were also collected from 6 sites on Robinson Ridge (site RR) and the Bailey (sites RS, A2 and A3) and Clark peninsulas (C1 and C2) (Supplementary Fig. 1) between 2005 and 2013 during the austral summer season, and their ages were determined using radiocarbon dating (see below). As Antarctic moss turfs grow in a highly compact manner (approximately 20 shoots within 0.5 mm of one another21), samples were collected as a 1 cm2 ‘core’, with the assumption that shoot sections from within the core experience near-identical growth conditions and would therefore be of a similar age12. Cores were extracted from the moss turf, air-dried to a constant mass at 60 °C and frozen at −20 °C for transport back to Australia.

Determination of relative abundance

The relative abundance of each species in each quadrat and each year was modelled in R using the R2OpenBUGS Bayesian modelling package and similar methods introduced by Ashcroft et al.40 (see also the model and script for Fig. 2 in the Supplementary Information). The observed presence/absence data from the nine pinches in each quadrat were modelled using a binomial distribution with probability Py,s,q and number of samples n = 9. The probabilities were allowed to vary according to year (y), species (s) and quadrat (q), but were assumed to be nonlinearly related to the cover of the respective species. The Bayesian model used Markov chain Monte Carlo iteration to estimate the expected mean and 95% credible intervals (CIs) for the relative abundances of the three species and moribund moss given the observed samples, with the relative abundances constrained to sum to 100%. Note that C. purpureus was not detected before 2003 (see Fig. 1 and Supplementary Table 1; refs 24,36,41), but the 95% CI still indicates that the species could have been present but unobserved; hence, the modelled abundance shows this species possibly occurring at low levels in 2000 (Fig. 2).

Digital evaluation of moss health

Quadrats were digitally photographed in all years except 2000 and 2011 (Supplementary Table 2). Digital red–green–blue (RGB) composite image analysis of the 25 cm × 25 cm permanent vegetation-monitoring quadrats was used to assess the quadrat percentage cover of moss, which was categorized as healthy (green), stressed (red–brown) or moribund (grey–black; lacking pigments and photosynthetically inactive). The physiological basis underlying this categorization is that when moss resources are invested in photosynthesis and growth, plant tissues are high in chlorophyll content and appear green, whereas mosses under drought and/or light stress produce more protective pigments, including anthocyanins, flavonoids and carotenoids, which give the plants a red or brown appearance39,42,43. These pigments have protective roles, including as sunscreens and antioxidants, which are especially important in high-radiation environments such as Antarctica. The extent to which the plants appear green or red depends on the relative proportions of these pigments in the cells and their cellular locations26,39. When mosses become moribund, their tips lose all pigment content and appear grey or black in colour. Although they may essentially be dead, if any viable cells remain they can regrow; thus, these mosses are designated moribund.

The analysis was conducted using semi-automated object-based image analysis (OBIA) methods, based on RGB colour ratios. To estimate cover, the photos were first imported into the ArcGIS geographical information system (ESRI) for preprocessing. They were georeferenced to their corresponding Global Positioning System locations and consequently resized to the correct quadrat size of 25 cm × 25 cm. Masking digital polygons were manually delineated to define the extent of the quadrat and mask out large rocks within each image. The preprocessed images and digital masks were then imported into the eCognition (Definiens Imaging) software package for OBIA and classification. Optimal settings for image object segmentation were determined using the Estimation of Scale Parameter (ESP) tool44. Multiresolution image segmentation was performed using a scale parameter of 33, with parameters of colour/shape set at 0.9/0.1 and smoothness/compactness at 0.5/0.5. The subsequent image classifier, which used a set of rules based on RGB and hue–saturation–intensity digital value thresholds, was applied to separate moss turf within the quadrats into healthy, stressed and moribund moss categories. The eCognition OBIA classification of selected images was compared with a manual classification performed by the operator, resulting in an overall accuracy of 84%. Average moss cover was >94 and >85% in the quadrats at ASPA 135 and Robinson Ridge, respectively, with a small proportion of other abiotic cover (that is, rock, snow, ice or water) and lichens. Abiotic and lichen cover (total other categories: 1–6% at ASPA 135 and 12–16% at Robsinson Ridge) did not change significantly for either site throughout the decade. For this reason, only the moss cover data (healthy, stressed and moribund) are presented in this paper (Fig. 3).

Dating and stable carbon isotope analysis

Mosses lay down sequential carbon signals reflecting the environmental conditions during growth13,45. Intact moss shoots can thus be accurately dated using ‘bomb-peak’ radiocarbon dating, where a peak in recent atmospheric radiocarbon occurred when large quantities of artificially produced 14C were injected into the atmosphere during the period of nuclear testing mostly in the late 1950s and early 1960s12,46. Data from four previously dated C. purpureus cores12 have also been included, which were remodelled to capture recent updates to atmospheric CO2 values47 as changes to atmospheric CO2 influence core δ13C through time48.

Samples were identified to species level. Cores were prepared for accelerator mass spectrometry (AMS) radiocarbon dating and δ13Cgraphite (a part of the 14C dating process used for fractionation correction) analysis at the Australian Nuclear Science and Technology Organisation, as described by Clarke et al.12. Each core was cut into 2 or 3 mm longitudinal segments using clean metal instruments to avoid carbon contamination. The moss samples were pretreated in hot 2 M HCl solution to remove possible carbonate contamination before being combusted to CO2 and then converted to graphite for AMS 14C analysis using the STAR Facility at the Australian Nuclear Science and Technology Organisation12. A small portion of the graphite was used for stable isotopic carbon (δ13C) determination to correct for isotopic fractionation of 14C using an elemental analyser/isotope-ratio mass spectrometer. To obtain accurate calendar ages for the consecutive moss segments of a particular core, their obtained 14C content, referred to as the percentage of modern carbon (pMC), was modelled using a Bayesian analysis approach—the ‘Simple Sequence’ deposition model of the OxCal version 4.2 programme49.

This approach calibrates their ages against our constructed atmospheric 14C curves and models the moss ages based on the assumption that subsequent moss segments within one core grew in a chronological manner. The constructed atmospheric 14C curves consist of Southern Hemisphere summer 14C data for the pre-bomb (Southern Hemisphere Calibration SHCal13; ref. 50) and post-bomb (Southern Hemisphere zone 1-2; ref. 51) extended beyond 2011 by exponential extrapolation. Each moss sample might span more than one year of growth, as indicated by the fact that the maximum pMC values for each long shoot core were lower than the annual bomb‐peak value of the Southern Hemisphere zone 1-2 data of ~166 pMC51. This time integration of moss samples was therefore taken into account when building the constructed atmospheric 14C curves (see ref. 12). The dataset of modern 14C ages of the moss cores depicted in Fig. 4 is included as Supplementary Fig. 2.

In addition to recording their age in their cells, as described above, mosses have the potential to record past bioavailable water climates via assimilation of stable carbon isotopes (12C and 13C) from the atmosphere. The enzyme RuBisCO favours the fixation of lighter 12CO2 during photosynthesis52,53,54,55. Due to their lack of stomata, there is no diffusional limitation imposed by stomatal control and so mosses fix more diffused 12CO2 than 13CO2 into sugars and cellulose when the leaf surface is dry. In wetter environments, when the moss surface is submerged in a layer of water (see Fig. 1), RuBisCO becomes CO2 limited and discriminates less against 13CO2; thus, more of the heavier stable carbon isotope is incorporated as 13Ccellulose54. The change (δ) in the ratio between fixed carbon-12 and carbon-13 (δ13Ccellulose) indicates the water environment (dry versus wet) and correlates with the measured δ13Cgraphite required for AMS 14C correction (above; see Bramley-Alves et al.13 for more information). Correction of this moss shoot δ13Cgraphite for temporal changes in atmospheric δ13C due to fossil fuel emissions was undertaken according to Clarke et al.12.

To determine the trends in immediate water environment for each moss core, the rate of change in δ13Cgraphite per year of dated intact moss shoot core sections was obtained using linear regression analysis on all data since 1960. The resultant slopes of these regressions are presented in Fig. 4. We restricted our analysis to dates after 1960 to focus on recent changes and because dates before the bomb-peak could not be estimated as precisely.

Analysis of local and global climate trends

Meteorology data, including daily minimum and maximum temperatures, precipitation and average and maximum wind speed, were obtained from the BOM for Casey (station 300017, 1990–2017), with earlier observations from the nearby Wilkes (station 300003, 1961–1969) and Casey Tunnel stations (station 300006, 1970–1989) (Supplementary Fig. 1). Data for the SAM15,17 were obtained from the British Antarctic Survey. The data for each year were taken as the average for the austral summer (December of the preceding year through January and February). As long-term meteorological trends can be confounded by both documented and undocumented changes in equipment, land use and location56, we modelled the time series using four linear segments, with three change points to cater for potential discontinuities. Each linear segment had to be at least five years long, to avoid overfitting to individual years, and could vary in both slope and intercept. The years for the change points could change in each iteration. Uninformed priors were used. When averaged over 20,000 iterations, this produced a 95% CI for the overall climate trend as well as the probability that a change point occurred in each year57. As the change points were not always in the same years, the mean lines were a nonlinear fit to the data. Correlations between regional meteorological and continent-wide SAM data were analysed conservatively by station location, in case local changes in instrument location affected records. None of the time series had high probabilities of discontinuities at the same time as station moves (1969 and 1989), so we assumed shifts represented actual shifts in climate rather than station biases.

While the results are only presented for degree days and mean wind speeds, minimum and maximum temperatures exhibited similar trends to degree days, and maximum wind speeds exhibited similar trends to mean wind speeds. Precipitation could not be analysed as there are problems recording snow due to the strong winds and frequent occurrences of blown snow.

We analysed and present the correlations for SAM versus degree days and mean wind speeds as factors that potentially drive the drying trend (Fig. 5). We acknowledge that these correlations with SAM do not prove causation, but there is previous literature that supports SAM leading to increased wind speeds and cooler maximum temperatures1,3,16,17. We also examined correlations between meteorological observations and ozone observations, zonal wave 3 and its Australian zone variant, but the results are not presented as trends were weaker, these factors were not as well supported by previous literature, and we preferred to test a priori hypotheses rather than examine post hoc correlations.

Data availability

Datasets are publicly available from the Australian Antarctic Data Centre (AADC) at https://doi.org/10.4225/15/59c999a4c2145.

Additional information

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.


  1. 1.

    Doran, P. T. et al. Antarctic climate cooling and terrestrial ecosystem response. Nature 415, 517–520 (2002).

  2. 2.

    Convey, P. et al. Antarctic climate change and the environment. Antarct. Sci. 21, 541–563 (2009).

  3. 3.

    Turner, J. et al. Antarctic climate change and the environment: an update. Polar Rec. 50, 237–259 (2014).

  4. 4.

    Robinson, S. A., Wasley, J. & Tobin, A. K. Living on the edge—plants and global change in continental and maritime Antarctica. Glob. Change Biol. 9, 1681–1717 (2003).

  5. 5.

    Guglielmin, M., Fratte, M. D. & Cannone, N. Permafrost warming and vegetation changes in continental Antarctica. Environ. Res. Lett. 9, 045001 (2014).

  6. 6.

    Brabyn, L. et al. Quantified vegetation change over 42 years at Cape Hallett, East Antarctica. Antarct. Sci. 18, 561–572 (2006).

  7. 7.

    Turner, J. et al. Absence of 21st century warming on Antarctic Peninsula consistent with natural variability. Nature 535, 411–415 (2016).

  8. 8.

    Bromwich, D. H. et al. Central West Antarctica among the most rapidly warming regions on Earth. Nat. Geosci. 6, 139–145 (2013).

  9. 9.

    Parnikoza, I. et al. Current status of the Antarctic herb tundra formation in the central Argentine Islands. Glob. Change Biol. 15, 1685–1693 (2009).

  10. 10.

    Amesbury, M. J. et al. Widespread biological response to rapid warming on the Antarctic Peninsula. Curr. Biol. 27, 1616–1622 (2017).

  11. 11.

    Hill, P. W. et al. Vascular plant success in a warming Antarctic may be due to efficient nitrogen acquisition. Nat. Clim. Change 1, 50–53 (2011).

  12. 12.

    Clarke, L. J., Robinson, S. A., Hua, Q., Ayre, D. J. & Fink, D. Radiocarbon bomb spike reveals biological effects of Antarctic climate change. Glob. Change Biol. 18, 301–310 (2012).

  13. 13.

    Bramley-Alves, J., Wanek, W., French, K. & Robinson, S. A. Moss δ13C: an accurate proxy for past water environments in polar regions. Glob. Change Biol. 21, 2454–2464 (2015).

  14. 14.

    Hodgson, D. A. et al. Recent rapid salinity rise in three East Antarctic lakes. J. Paleolimnol. 36, 385–406 (2006).

  15. 15.

    Marshall, G. J. Trends in the Southern Annular Mode from observations and reanalyses. J. Clim. 16, 4134–4143 (2003).

  16. 16.

    Robinson, S. A. & Erickson, D. J. Not just about sunburn—the ozone hole’s profound effect on climate has significant implications for Southern Hemisphere ecosystems. Glob. Change Biol. 21, 515–527 (2015).

  17. 17.

    Marshall, G. J. & Thompson, D. W. J. The signatures of large-scale patterns of atmospheric variability in Antarctic surface temperatures. J. Geophys. Res. Atmos. 121, 3276–3289 (2016).

  18. 18.

    Abram, N. J. et al. Evolution of the Southern Annular Mode during the past millennium. Nat. Clim. Change 4, 564–569 (2014).

  19. 19.

    Pecl, G. T. et al. Biodiversity redistribution under climate change: impacts on ecosystems and human well-being. Science 355, eaai9214 (2017).

  20. 20.

    Convey, P. et al. The spatial structure of Antarctic biodiversity. Ecol. Monogr. 84, 203–244 (2014).

  21. 21.

    Wasley, J., Robinson, S. A., Lovelock, C. E. & Popp, M. Some like it wet—biological characteristics underpinning tolerance of extreme water stress events in Antarctic bryophytes. Funct. Plant Biol. 33, 443–455 (2006).

  22. 22.

    Lee, J. R. et al. Climate change drives expansion of Antarctic ice-free habitat. Nature 547, 49–54 (2017).

  23. 23.

    Lenné, T., Bryant, G., Hocart, C. H., Huang, C. X. & Ball, M. C. Freeze avoidance: a dehydrating moss gathers no ice. Plant Cell Environ. 33, 1731–1741 (2010).

  24. 24.

    Wasley, J. et al. Bryophyte species composition over moisture gradients in the Windmill Islands, East Antarctica: development of a baseline for monitoring climate change impacts. Biodiversity 13, 257–264 (2012).

  25. 25.

    Melick, D. & Seppelt, R. Vegetation patterns in relation to climatic and endogenous changes in Wilkes Land, continental Antarctica. J. Ecol. 85, 43–56 (1997).

  26. 26.

    Malenovský, Z., Turnbull, J. D., Lucieer, A. & Robinson, S. A. Antarctic moss stress assessment based on chlorophyll content and leaf density retrieved from imaging spectroscopy data. New Phytol. 208, 608–624 (2015).

  27. 27.

    Williamson, C. E. et al. Solar ultraviolet radiation in a changing climate. Nat. Clim. Change 4, 434–441 (2014).

  28. 28.

    Assessment for Decision-Makers: Scientific Assessment of Ozone Depletion: 2014 Report No. 56 (World Meteorological Organization, 2014).

  29. 29.

    Villalba, R. et al. Unusual Southern Hemisphere tree growth patterns induced by changes in the Southern Annular Mode. Nat. Geosci. 5, 793–798 (2012).

  30. 30.

    Weimerskirch, H., Louzao, M., De Grissac, S. & Delord, K. Changes in wind pattern alter albatross distribution and life-history traits. Science 335, 211–214 (2012).

  31. 31.

    Gooseff, M. N. et al. Decadal ecosystem response to an anomalous melt season in a polar desert in Antarctica. Nat. Ecol. Evol. 1, 1334–1338 (2017).

  32. 32.

    Dunn, J. L. & Robinson, S. A. Ultraviolet B screening potential is higher in two cosmopolitan moss species than in a co-occurring Antarctic endemic moss: implications of continuing ozone depletion. Glob. Change Biol. 12, 2282–2296 (2006).

  33. 33.

    Lovelock, C. E. & Robinson, S. A. Surface reflectance properties of Antarctic moss and their relationship to plant species, pigment composition and photosynthetic function. Plant Cell Environ. 25, 1239–1250 (2002).

  34. 34.

    Robinson, S. A., Wasley, J. & King, D. Indicator 72—Windmill Islands Terrestrial Vegetation Dynamics (Australian Antarctic Data Centre, 2009); https://go.nature.com/2xm08l7

  35. 35.

    Australia State of the Environment 2011: Independent Report to the Australian Government Minister for Sustainability, Environment, Water, Population and Communities 477–565 (State of the Environment 2011 Committee, 2011).

  36. 36.

    Lewis Smith, R. I. Plant community dynamics in Wilkes Land, Antarctica. Proc. NIPR Symp. Polar Biol. 3, 229–244 (1990).

  37. 37.

    Emslie, S. D. & Woehler, E. J. A 9000-year record of Adélie penguin occupation and diet in the Windmill Islands, East Antarctica. Antarct. Sci. 17, 57–66 (2005).

  38. 38.

    Robinson, S. A., Wasley, J., Popp, M. & Lovelock, C. E. Desiccation tolerance of three moss species from continental Antarctica. Funct. Plant Biol. 27, 379–388 (2000).

  39. 39.

    Waterman, M. J. The What and Where of Ultraviolet Protective Mechanisms in Antarctic Mosses. PhD thesis, Univ. Wollongong (2015).

  40. 40.

    Ashcroft, M. B. et al. Moving beyond presence and absence when examining changes in species distributions. Glob. Change Biol. 23, 2929–2940 (2017).

  41. 41.

    Selkirk, P. M. & Seppelt, R. D. Species distribution within a moss bed in Greater Antarctica. Symp. Biol. Hungarica 35, 279–284 (1987).

  42. 42.

    Post, A. Photoprotective pigment as an adaptive strategy in the Antarctic moss Ceratodon purpureus. Polar Biol. 10, 241–245 (1990).

  43. 43.

    Waterman, M. J. et al. Antarctic moss biflavonoids show high antioxidant and ultraviolet-screening activity. ‎J. Nat. Prod. 80, 2224–2231 (2017).

  44. 44.

    Drǎguţ, L., Tiede, D. & Levick, S. R. ESP: a tool to estimate scale parameter for multiresolution image segmentation of remotely sensed data. Int. J. Geogr. Inf. Sci. 24, 859–871 (2010).

  45. 45.

    Royles, J. et al. Plants and soil microbes respond to recent warming on the Antarctic Peninsula. Curr. Biol. 23, 1702–1706 (2013).

  46. 46.

    Hua, Q. Radiocarbon: a chronological tool for the recent past. Quat. Geochronol. 4, 378–390 (2009).

  47. 47.

    Rubino, M. et al. A revised 1000 year atmospheric δ13C-CO2 record from Law Dome and South Pole, Antarctica. J. Geophys. Res. Atmos. 118, 8482–8499 (2013).

  48. 48.

    Dan, Y. The paper trail of the 13C of atmospheric CO2 since the industrial revolution period. Environ. Res. Lett. 6, 034007 (2011).

  49. 49.

    Bronk Ramsey, C. Bayesian analysis of radiocarbon dates. Radiocarbon 51, 337–360 (2009).

  50. 50.

    Hogg, A. G. et al. SHCal13 Southern Hemisphere calibration, 0–50,000 years cal BP. Radiocarbon 55, 1889–1903 (2013).

  51. 51.

    Hua, Q., Barbetti, M. & Rakowski, A. Z. Atmospheric radiocarbon for the period 1950–2010. Radiocarbon 55, 2059–2072 (2013).

  52. 52.

    Williams, T. & Flanagan, L. Effect of changes in water content on photosynthesis, transpiration and discrimination against 13CO2 and C18O16O in Pleurozium and Sphagnum. Oecologia 108, 38–46 (1996).

  53. 53.

    Rice, S. Variation in carbon isotope discrimination within and among Sphagnum species in a temperate wetland. Oecologia 123, 1–8 (2000).

  54. 54.

    Rice, S. K. & Giles, L. The influence of water content and leaf anatomy on carbon isotope discrimination and photosynthesis in Sphagnum. Plant Cell Environ. 19, 118–124 (1996).

  55. 55.

    Farquhar, G., Ehleringer, J. & Hubick, K. Carbon isotope discrimination and photosynthesis. Annu. Rev. Plant. Physiol. Plant. Mol. Biol. 40, 503–537 (1989).

  56. 56.

    Ducré-Robitaille, J.-F., Vincent, L. A. & Boulet, G. Comparison of techniques for detection of discontinuities in temperature series. Int. J. Climatol. 23, 1087–1101 (2003).

  57. 57.

    Ruggieri, E. A Bayesian approach to detecting change points in climatic records. Int. J. Climatol. 33, 520–528 (2013).

Download references


The authors wish to thank D. Bergstrom, Z. Malenovský, A. Nydahl, J. Dunn, A. Lucieer and other Australian National Antarctic Research Expedition expeditioners for assistance in the field, A. Netherwood for production of Fig. 1, and B. Raymond and A. Constable for providing feedback on the manuscript. Funding was provided by the Australian Research Council (DP110101714 and DP180100113), Antarctic Science Grants 1313, 3129, 3042 and 4046, and Australian Institute of Nuclear Science and Engineering grants 05142P and 06155. We acknowledge financial support from the Australian Government for the Centre for Accelerator Science at ANSTO through the National Collaborative Research Infrastructure Strategy and the University of Wollongong’s Global Challenges Program as part of the Sustaining Coastal and Marine Zones challenge. J.W., L.J.C., J.B.-A., M.J.W. and D.H.K. were supported by Australian Postgraduate Awards/Research Training Program scholarships. M.J.W. also received an Australian Institute of Nuclear Science and Engineering postgraduate award (grant ALNSTU2110).

Author information

Author notes

    • Jane Wasley
    •  & Laurence J. Clarke

    Present address: Antarctic Conservation and Management Program, Australian Antarctic Division, Kingston, Tasmania, Australia

    • Rebecca E. Miller

    Present address: School of Ecosystem and Forest Sciences, University of Melbourne, Melbourne, Victoria, Australia

    • Ellen Ryan-Colton

    Present address: Research Institute for the Environment and Livelihoods, Charles Darwin University, Alice Springs, Northern Territory, Australia

    • Laurence J. Clarke

    Present address: Antarctic Climate and Ecosystems Cooperative Research Centre, University of Tasmania, Hobart, Tasmania, Australia

  1. These authors contributed equally: Sharon A. Robinson, Diana H. King, Jessica Bramley-Alves, Melinda J. Waterman, Michael B. Ashcroft, Jane Wasley.


  1. Centre for Sustainable Ecosystem Solutions, School of Biological Sciences, University of Wollongong, Wollongong, New South Wales, Australia

    • Sharon A. Robinson
    • , Diana H. King
    • , Jessica Bramley-Alves
    • , Melinda J. Waterman
    • , Michael B. Ashcroft
    • , Jane Wasley
    • , Johanna D. Turnbull
    • , Rebecca E. Miller
    • , Ellen Ryan-Colton
    • , Taylor Benny
    • , Kathryn Mullany
    •  & Laurence J. Clarke
  2. Global Challenges Program, University of Wollongong, Wollongong, New South Wales, Australia

    • Sharon A. Robinson
  3. Australian Nuclear Science and Technology Organisation, Sydney, New South Wales, Australia

    • Linda A. Barry
    •  & Quan Hua
  4. Antarctic Climate and Ecosystems Cooperative Research Centre, University of Tasmania, Hobart, Tasmania, Australia

    • Laurence J. Clarke


  1. Search for Sharon A. Robinson in:

  2. Search for Diana H. King in:

  3. Search for Jessica Bramley-Alves in:

  4. Search for Melinda J. Waterman in:

  5. Search for Michael B. Ashcroft in:

  6. Search for Jane Wasley in:

  7. Search for Johanna D. Turnbull in:

  8. Search for Rebecca E. Miller in:

  9. Search for Ellen Ryan-Colton in:

  10. Search for Taylor Benny in:

  11. Search for Kathryn Mullany in:

  12. Search for Laurence J. Clarke in:

  13. Search for Linda A. Barry in:

  14. Search for Quan Hua in:


S.A.R., D.H.K., J.B.-A., M.J.W., J.W., J.D.T., E.R.-C. and L.J.C. conceived the experiments. J.B.-A., J.W., J.D.T., S.A.R., R.E.M., E.R.-C. and L.J.C. performed the fieldwork. D.H.K. performed the image analysis. D.H.K., J.B.-A., J.W., J.D.T., T.B. and K.M. processed the moss microsamples. L.J.C., S.A.R., J.B.-A. and M.J.W. identified samples to species level. M.J.W., J.B.-A., L.J.C., L.A.B. and Q.H. performed the dating and isotope analysis. M.B.A., D.H.K., M.J.W., J.B.-A. and Q.H. analysed the data. S.A.R., M.B.A., M.J.W., D.H.K., J.B.-A., J.W., J.D.T., R.E.M. and Q.H. co-wrote the manuscript.

Competing interests

The authors declare no competing interests

Corresponding author

Correspondence to Sharon A. Robinson.

Supplementary information

  1. Supplementary Information

    Supplementary tables S1–S2, Supplementary figures 1–3, Supplementary references

  2. Fig2_Model.txt

    R2OpenBUGS Bayesian model used by the R script (Fig2_script.txt) to estimate the change in relative abundance of three moss species and moribund (dead or dying) moss over time

  3. Fig2_script.txt

    The relative abundance of each species in each quadrat and each year was modelled in R using the R2OpenBUGS Bayesian modelling package (see Fig2_Model.txt) and this R script (Fig2_script.txt)

About this article

Publication history




Issue Date



Further reading