Temporal dynamics in alpine snowpatch plants along a snowmelt gradient explained by functional traits and strategies

Alpine snowpatches are characterised by persistent snow cover, short growing seasons and periglacial processes, which has resulted in highly specialised plant communities. Hence, these snowpatch communities are among the most threatened from climate change. However, temporal dynamics in snowpatch microclimate and plant composition are rarely explored, especially in the marginal alpine environments of Australia. Seven snowpatches were categorised into early, mid and late snowmelt zones based on growing season length, with soil temperatures recorded from 2003 to 2020 and plant composition surveyed in 84 1 m2 quadrats in 2007, 2013 and 2020. Microclimate, species diversity, plant cover and composition, along with community-weighted trait means and plant strategies were assessed to understand snowpatch dynamics in response to climate change. We found that growing season length and temperatures have increased in late melt zones, while changes were less consistent in early and mid melt zones. There were few changes in species diversity, but increases in graminoids and declines in snowpatch specialists in mid and late melt zones. Community-weighted plant height, leaf area and leaf weight also increased, particularly in mid and late melt zones, while plant strategies shifted from compositions of ruderal-tolerant to stress-tolerant. Here, we show that snowpatch communities are rapidly changing in response to longer growing seasons and warmer temperatures, with the greatest changes occurring where snow persists the longest. The results highlight the climate-induced loss of defining biotic and abiotic characteristics of snowpatches, as temporal convergence of compositions along snowmelt gradients threatens the distinctiveness of snowpatch plant communities.


Introduction
Alpine ecosystems are directly governed by low temperatures with short growing seasons due to seasonal snow cover (Körner 2003). Consequently, they are among the ecosystems most threatened globally from climate change, as temperatures increase in many alpine regions resulting in declines in snow cover (Guisan et al. 2019;Huss et al. 2017). As the duration of snow cover is a strong environmental filter for plant composition (Körner 2003), the threat of climate change is amplified for cryophilic plant communities that are maintained by periglacial and nivation processes (Björk and Molau 2007). In topographical depressions on lee aspects of mountain ridgelines where snow accumulates throughout the winter, snowpatches may persist for months after the general thaw, supporting distinctive and highly specialised snowpatch (synonymous with 'snowbed' and 'snowbanks') species and communities (Billings and Bliss 1959;Björk and Molau 2007;Green and Pickering 2009a). Although research on snowpatch plant communities remains sparse (Verrall and Pickering 2020), they are considered among the most vulnerable alpine ecosystems to climate change (Gritsch et al. 2016;Matteodo et al. 2016). This includes Australia where alpine environments are already marginal and uncommon (Kirkpatrick et al. 2017;Pickering et al. 2014;Williams et al. 2015). Alpine ecosystems cover less 0.02% of Australia, as they are limited to a narrow elevation range with the bioclimatic treeline is only a few hundred metres below the highest peaks, and no nival zone (Green and Stein 2015). Snowpatch plant communities meet the criteria for listing as endangered under the International Union for Conservation of Nature (IUCN) Red List as they are particularly rare in Australia, occupying less than 1% of the total alpine area but provide important refugia for dwarf alpine plant species in the face of climate change (Williams et al. 2015). The most persistent of these snowpatches occur in the Australian Alps near Mount Kosciuszko/Kunama Namadgi, Australia's tallest mountain at 2228 m above sea level (a.s.l) (Green and Pickering 2009a). Snow cover, depth and thaw date in the Kosciuszko alpine area are variable (Duus 1992;Nicholls 2005), with a marginal winter snowpack usually forming by late May and becoming discontinuous in October (Costin et al. 2000). However, snowpatches persist in semi-regular positions in nivation hollows on the south to south-east of ridgelines where snow-laden north-westerly winds load snow throughout the winter, and ablation is abated by limited insolation in the spring and summer (Edmonds et al. 2006). Historically, the largest of these snowpatches accumulated depths of 30 m and underwent appreciable nivation, persisting for several years as firn (Costin et al. 1964). However, snow cover is declining in the Kosciuszko alpine area as conditions warm with climate change (McGowan et al. 2018;Nicholls 2005;Sánchez-Bayo and Green 2013).
Climate change is already affecting abiotic conditions associated with snowpatches, with longer growing seasons due to reductions in the extent and persistence of snow cover in the Australian Alps (Edmonds et al. 2006;Green and Pickering 2009a), as well as in other alpine regions (Huss et al. 2017;Woo and Young 2014). To assess potential compositional changes in vegetation associated with snowpatches in response to climate change, a range of methods have been used such as simulated environmental changes (Bernareggi et al. 2015;Kudo et al. 2010;Sandvik et al. 2004), as well as space-for-time substitutions along snowmelt gradients (Carbognani et al. 2012;Good et al. 2019;Schöb et al. 2009). However, there are few long-term ecological monitoring initiatives focussing on compositional changes in snowpatch vegetation, such as those in Norway (Sandvik and Odland 2014), Italy , Japan (Amagai et al. 2018), the Australian Alps (Pickering et al. 2014) and in Tasmania, Australia (Kirkpatrick et al. 2017).
Commonly, lengthening growing seasons have resulted in changing plant composition Pickering et al. 2014;Sandvik and Odland 2014). In the Australian Alps, for instance, there was increased cover of the tall tussock graminoid, Poa costiniana, between surveys in 2007 and 2013, as well as increasing species richness where snow persists the longest (Pickering et al. 2014). There were also declines in the presence and cover of snowpatch specialists (Pickering et al. 2014), as has occurred in other alpine regions Sandvik and Odland 2014). In contrast, in the more maritime alpine environments of Tasmania, Australia snowpatch specialists remained stable, although the cover of cushion plants and tall shrubs increased in snowpatches between 1983(Kirkpatrick et al. 2017. As composition changed in the Australian Alps, there were also increases in community-weighted trait mean plant height and specific leaf area (Pickering et al. 2014). Together, studies from a range of alpine areas indicate that snowpatches may now be in disequilibrium with the changing climate with diverse responses including increasing competitive exclusion and succession by adjacent plant species and communities (Matteodo et al. 2016;Pickering et al. 2014;Schöb et al. 2009;Venn et al. 2011). They highlight the importance of long-term monitoring to understand how snowpatches, which were previously distinctive communities, are affected by changing climates (Bergstrom et al. 2021;Williams et al. 2015).
However, temporal dynamics in functional traits and composition are rarely assessed in snowpatches, even though they are known to vary considerably along snowmelt gradients (Good et al. 2019;Pickering et al. 2014). Functional traits reflect plant ecological strategies and serve as strong predictors of community dynamics in response to climate change (Enquist et al. 2015;Guittar et al. 2016). Commonly measured traits in alpine plants include plant height, leaf area and leaf mass (Pickering et al. 2014;Testolin et al. 2021;Tonin et al. 2019;Venn et al. 2011). Plant height indicates species' overall competitive ability for light (Falster and Westoby 2003), but is also associated with flowering phenology and seed dispersal (Sun and Frelich 2011;Thomson et al. 2011), and can correlate with above-and belowground biomass and leaf size (Cornelissen et al. 2003). Biomass allocation via leaf area and mass are also traits intrinsically linked to competitive light capture and drought tolerance (Yin et al. 2019), with important implications for water balance and leaf energy budgets (Cornelissen et al. 2003). Generally, alpine plants have relatively small plant height, leaf area and mass compared to many other species, as they are subject to the extreme cold, wind, drought and high radiation (Körner 2003;Testolin et al. 2021). However, it remains unclear how functional composition is responding with declining snow cover and lengthening growing seasons, including in the Australian Alps.
Since snowpatches are characterised by cryophilic, dwarf plant communities with distinguishing functional compositions that only occur within narrow microclimatic niches with short growing seasons, the loss of these defining characteristics may signal that snowpatches are approaching collapse via replacement by a novel ecosystem (Bergstrom et al. 2021;Pickering et al. 2014;Williams et al. 2015). Here, we use long-term ecological monitoring to assess the changes in the microclimate and plant composition of snowpatches located in the Australian Alps, where some changes have already been documented (Pickering et al. 2014). We hypothesise that there will be differences in plant community composition and growing season mean temperatures and length (microclimate) along snowmelt gradients, but the greatest temporal changes would occur where snow persists longest. Furthermore, we hypothesise that there will be an increase in generalists and subsequent decrease in snowpatch specialists, with an increase of community-weighted trait means for plant height and leaf mass leading to the loss of distinctive compositions along snowmelt gradients as plants are experiencing warmer temperatures and longer growing seasons.

Study site and snowpatch characteristics
A long-term ecological monitoring initiative was established in 2003 to assess seven of the most persistent snowpatches in Australian Alps Pickering 2009a, 2009b;Pickering et al. 2014). These snowpatches occur above 2000 m a.s.l. along a 10 km section of the Main Range near Mount Kosciuszko/Kunama Namadgi (2228 m a.s.l.), in Australia's largest and most diverse contiguous alpine area (Fig. 1a). This area experiences a mild, mid-latitude alpine climate where mean annual air temperature is 4.8 °C and mean annual precipitation is 1274.8 mm, based on records over the past 30 years (1991-2021) from the closest weather station (Thredbo AWS 071,032) located 7.84 km south (180.71°) of these snowpatches at an elevation of 1957 m a.s.l (BoM 2022). The snowpatches in this area are characterised by short growing seasons ranging from 80 to  (Pickering et al. 2014), a steady supply of meltwater and constant disturbance from periglacial and nivation processes such as gelifluction, abrasion and frost creep (Costin 1954). Here, transitions in plant communities reflect varying abiotic conditions along topographical and snowmelt gradients, with differences in rock cover, soil depth, nutrients as well as microclimatic heterogeneity including varying growing season length and temperature Pickering 2009a, 2009b). In the centre of snowpatches where snow persists the longest (~ late melt zone), skeletal soils support sparse vegetation dominated by Snowpatch Feldmark (Coprosma-Colobanthus alliance) species (Costin 1954) (Fig. 1b). Snowpatch Herbfields (Plantago-Neopaxia alliance) occur directly downslope from the centre of snowpatches where snow does not persist as long (~ mid melt zone), and soil depth, moisture and plant cover increased (Costin 1954;Green and Pickering 2009b). In the areas where snow melts first further downslope (~ early melt zone), species characteristic of Tall Alpine Herbfields (Celmisia-Poa alliance) occur on deep alpine humus soils (Costin 1954;Green and Pickering 2009b). Snowpatch Herbfield and Snowpatch Feldmark plant communities are listed as critically endangered by the New South Wales Government, as they are restricted to only the most persistent snowpatches in the Australian Alps (Duretto 2019a(Duretto , 2019b.

Sampling design
Since snowpatch ecosystem function is strongly correlated to soil temperature, it is commonly used to represent the microclimate experienced by dwarf snowpatch plant communities (Björk and Molau 2007;Green and Pickering 2009b). Across the seven snowpatches (hereafter referred to as "study area"), 29 soil temperature loggers (Tinytag Plus 2) buried at a depth of 75 mm were used to monitor microclimatic conditions along snowmelt gradients, with soil temperatures recorded every two hours from 1/6/2003 at the beginning of the winter until 31/5/2020 at the end of the autumn (17 years) (see Green and Pickering 2009b). These loggers have a reading range of − 40 °C to + 85 °C using an internally mounted 10 K negative temperature coefficient thermistor with a reading resolution of 0.01 °C and ± 0.5 °C accuracy at 0 °C. The distribution of these loggers was determined by the zonation of plant communities present in each snowpatch, which are determined by temporal sequences in snow duration and are zoned along topographical gradients (Costin 1954). Loggers were placed in the centre of three plant community zones with two transitional zones present in 2003: (A) Snowpatch Feldmark, (C) Snowpatch Herbfield, (E) Tall Alpine Herbfield with (B) and (D) transitioning between these communities. However, not all snowpatches contained all five zones (Table 1). Then, the distribution of these loggers provided the framework for the establishment of 84.1 m 2 permanently marked vegetation monitoring quadrats across the study area using stratified random sampling in 2007 (see Green and Pickering 2009b).
For each zone present in each snowpatch, three quadrats were randomly allocated at the same elevation across the slope as soil temperature loggers (i.e. loggers were not within quadrats), where the mean distance between quadrats was 4.5 ± 1.1 m. Plant composition was recorded in the 84.1 m 2 quadrats within study area in March 2007, March 2013 and February 2020 for all vascular species and bryophytes. Species richness and overlapping plant Table 1 Physical characteristics and zones sampled at each of the seven snowpatches-Twynam Cirque (TWY), Blue Lake North (BLN), Blue Lake South (BLS), Club Lake Cirque (CLC), Mawson Cirque (MAW), Kosciuszko North East (KNE) and Cootapatamba Cornice (CTB) Vegetation zones were of increasing vegetation height and cover from the centre of the snowpatch (A) to Tall Alpine Herbfield downslope of snowpatches (E). These vegetation zones were then reclassified into early, mid and late melt zones based on 1 °C growing season length obtained from soil temperature data, with the number of 1 m 2 vegetation quadrats sampled (n) in each survey (2007,2013,2020)  cover (shadow of each species at solar zenith) was visually assessed to 0.5% intervals using a 10 × 10 cm gridded 1 m 2 quadrat where each cell represents 1% cover, following methods adapted from Pauli et al. (2015) and earlier assessments (Green and Pickering 2009b;Pickering et al. 2014). Species taxonomy were recorded according to Costin et al. (2000) but then updated to the latest taxonomy according to the New South Wales Flora Online. All loggers and quadrats were permanently marked with metal pegs where soil depth permitted, with star pickets installed in relatively stable geomorphic areas downslope of snowpatches to permit relocation via triangulation. With some pegs unable to be relocated in the 2013 and/or 2020 resurveys due to the downslope movement of soil, rock and snow, quadrats were reallocated using length and bearing measurements from pickets. Plant functional traits were also recorded from at least 10 individuals of each species during the peak of the 2009 growing season. Destructive sampling was conducted in sites adjacent to quadrats with similar topography and comparable duration of snow cover (Pickering et al. 2014;Venn et al. 2011). The traits measured for each vascular species were plant height (mm), leaf area (mm 2 ), leaf fresh weight (mg) and leaf dry weight (mg). These traits were selected based on individual plant and community functions (Venn et al. 2011). Plant height was recorded prior to destructive sampling as the shortest distance between the ground and the upper most photosynthetic material, and was cross referenced with values in the local flora (Costin et al. 2000). For leaf traits, 25 leaves including petiole were harvested from a minimum of 10 plants and placed into a cool and moist sealed bag while being transported to cold storage in a laboratory prior to processing. Then, leaves were scanned using LeafArea LeafArea (A. P. Askew, University of Sheffield, UK, downloadable from the Nucleo DiverSus toolbox) to quantify leaf area, and leaf fresh weight was recorded (Venn et al. 2011). To obtain leaf dry weight, leaves were weighed after oven drying at 70 ℃ for 48 h. Mean values were calculated from the 25 leaves for each species, but there were five species with no trait data recorded (Table S1), and they were mainly cryptogams that accounted for < 4.5% of total cover across all three surveys.

Data processing and analysis
For each soil temperature logger, growing season length was calculated as number of days in the year where daily mean soil temperature was above the 1 °C threshold, and there was diurnal variation in the temperature trace for each logger (Bürli et al. 2021;Körner and Paulsen 2004). Since snow cover stabilises soil temperatures around 0 °C (Körner and Paulsen 2004), days below this 1 °C threshold and with stable temperature traces were not considered as part of the growing season. Using data from all days within the growing season, growing degree days (°d) were obtained by summing all daily temperatures and growing season mean temperatures were calculated. Yearly periods start on June 1st to coincide with the start of winter, and run until May 31st of the following year. Once growing season length was established, loggers and corresponding vegetation quadrats were allocated into three evenly split snowmelt categories (hereafter referred to as "melt zone") based on the first 10 years of monitoring (2003/04-2012/13): early melt (> 140 growing days; n = 33), mid melt (105 ≥ growing days ≥ 140; n = 27) and late melt (< 105 growing days, n = 24). To assess changes in the microclimatic conditions of growing season length, growing degree days and mean temperature, linear regressions (LMs) were conducted on melt zone means for each year and each melt zone via the baseline package "Stats" v4.1.0 (R Core Team 2022).
Plant composition data from quadrats were used to calculate species richness, Shannon index (Shannon 1948), 1-Simpson's index of dominance (Simpson 1949) and Sheldon evenness index (Sheldon 1969). Species' overlapping plant cover was summed to calculate plant cover, growth form cover (graminoids, forbs, cryptogams, shrubs) and species' association cover (snowpatch specialists, generalists), as outlined in Costin et al. (2000). From these cover data, community-weighted trait means including plant height (mm), leaf area (mm 2 ), leaf fresh weight (mg) and leaf dry weight (mg) were calculated from functional trait data (Pickering et al. 2014;Venn et al. 2011). Furthermore, functional traits allow the classification of species using the C-S-R triangle where species occur on continuums between three strategies: competitor (C), stress-tolerant (S) and ruderal-tolerant (R) (Grime 2006;Pierce et al. 2017). These three principal strategies in the CSR classification scheme are effectively represented by leaf area (mm 2 ), leaf dry matter content (%) and specific leaf area (mm 2 mg −1 ). Specifically, species with high dry matter leaf content (small, dense and tough leaves) are stress strategists, while species with high specific leaf area (small, soft leaves) are disturbance-tolerant, ruderal (R) strategists whereas species with large leaf area are competitor (C) strategists (Pierce et al. 2017). For each of the species, CSR proportions (%) were calculated from mean trait values using the tool 'Stratefy' developed by Pierce et al. (2017), but no species were found to be consistent with just C. Traits for each species were then used to determine community-weighted plant strategies for each quadrat by scaling the relative abundance (cover) of each species' competitor-stress-ruderal proportion using the 'StrateFy' tool (Good et al. 2019;Pierce et al. 2017).
To assess dynamics in species richness for the study area through time, and to determine if melt zones were predictors for species richness, a Generalised Linear Model (GLM) was conducted with "Melt Zone" and "Year" as fixed effects using quasipoisson distribution to account for over-dispersion and a log link function. To test if species' richness' dynamics were consistent across melt zones, quasipoisson GLMs were conducted for each melt zone with "Year" as the fixed effect. Diversity (Shannon, Dominance and Evenness index), cover (plant, graminoid, forb, cryptogam and snowpatch specialist cover) and functional trait (plant height, leaf area, leaf fresh and dry weight) data were observed to have beta distribution. Thus, data were parameterised from the original scale to the open interval (0, 1), following the protocol of Smithsen and Verkuilen (2005) and Damgaard and Irvine (2019) (see supplementary materials). After parameterisation, beta regression models were conducted with "Melt Zone" and "Year" as fixed effects using beta distribution and logit link function to assess dynamics across by means of the package "betareg" v3.1-4 (Zeileis et al. 2021). For all GLMs and beta regression models, F statistics and p-values were attained for fixed effects using the function "Anova" from package "car" v3.0-11 (Fox et al. 2021) and differences between levels of fixed effects were obtained via the package "emmeans" v1.6.3 (Lenth et al. 2021).
To assess variation in composition through time, Principal Components Analysis using variance-covariance matrices between groups were executed on cover data to visualise dynamics in species, growth form and species' association. Influential components were determined by Similarity Percentage Analysis (SIMPER) and visualised by proportional biplot vectors by means of the package "vegan" v2.5-7 (Oksanen et al. 2020). Furthermore, dynamics in plant strategies through time were visualised using ternary plots with Gaussian kernel density estimation heatmapping, and were generated with PAST v4.03 (Hammer et al. 2001). To determine if there had been compositional shifts through time, One-way Analysis of Similarity (ANOSIM) with "Year" as the fixed effect was conducted with a Bonferroni adjustment on pairwise differences between surveys by means of the package "vegan" v2.5-7 (Oksanen et al. 2020).

Microclimate
There was variation in microclimatic conditions across snowmelt zones in all 17 years (Fig. 2). However, over time, conditions within the late melt zone have become milder, with a significant annual increase in the growing season length by 2.16 days (r 2 = 0.262, p = 0.043), growing degree days by 25.40 (r 2 = 0.267, p = 0.041) and growing season mean temperature by 0.072 °C (r 2 = 0.254, p = 0.046) ( Table S2). In contrast, changes were less consistent for early and mid melt zones.

Diversity, cover and composition
A total of 67 plant species from 24 families were recorded across the three surveys of the seven snowpatches (Table S1). This included 43 forbs, 18 graminoids, four cryptogams and two shrubs, with 13 species considered snowpatch specialists. Although there was increasing species richness through time from 52 in 2007 to 57 in 2013 and then 67 in 2020, and there was clear difference across melt zones (F = 12.930, p < 0.001) (Table S3), there were few temporal shifts in diversity (Fig. 3). In the early melt zone between 2007 and 2013, Shannon index decreased (∆μ = 0.38 ± 0.11, d = 0.921, p = 0.044) and dominance increased (∆μ = 0.22 ± 0.05, d = 1.029, p = 0.046) but diversity stabilised in 2020 (Table S4). Dominance decreased in
Across the study area, species varied from stress to ruderal strategists, with only a few forbs with low cover classified as competitor-stress or competitor-ruderal (Table S1). A clear gradient of plant strategies was apparent among melt zones in 2007, with a stress-tolerant composition in the early melt zone, a ruderal-stress-tolerant composition in the mid melt zone and a ruderal-tolerant composition in late melt zone (Fig. 7). However, this changed over time with an overall shift in composition from ruderal-tolerant to stress-tolerant between 2007 and 2020 for the study area (R = 0.127, p < 0.001), early melt zone (R = 0.105, p = 0.002), mid melt zone (R = 0.102, p = 0.035) and late melt zone (R = 0.245, p = 0.002).

Vegetation dynamics
We show that plant diversity, cover and composition varied among melt zones and through time in some of the most persistent snowpatches in the Australian Alps. Our results support our hypotheses that changes were more pronounced where snow persists the longest, with Fig. 4 Dynamics in overlapping a plant cover, b graminoid cover, c forb cover, d) cryptogam cover, and e snowpatch specialists cover for the study area and for each melt zone for 2007, 2013 and 2020, with data points from each quadrat (colour-coded dots), distribution and density of data (colour-coded polygons), mean (black dot) and 95% confidence intervals (black line) shown. Asterisks indicate significant changes between years (*p < 0.05; **p < 0.01; ***p < 0.001) ◂ compositional differences driven by an increase in generalist and subsequent decrease in snowpatch specialists, leading to an increase of community-weighted trait means for plant height and leaf mass. Where previous studies concerning the response of snowpatch plant communities to climate change have either focussed on space-for-time assessments (Carbognani et al. 2012;Good et al. 2019;Schöb et al. 2009) or temporal changes between two sampling periods (Amagai et al. 2018;Sandvik and Odland 2014), this study presents novel insights about snowmelt-dependant vegetation responses across three sampling periods. In 2007, the early melt zone where conditions were mildest was characterised low diversity and by near complete plant cover consisting mainly of tall tussock-forming graminoids. The mid melt zone was more diverse including cryptogams, forbs and short graminoids, with several snowpatch specialists, with ~ 70% plant cover. In contrast, the late melt zone had just ~ 25% plant cover, mostly snowpatch specialists with the rest of the area mostly rock (Green and Pickering 2009b). These differences, in part, reflect how snow cover and growing season influence plant composition within snowpatches (Björk and Molau 2007;Good et al. 2019;Green and Pickering 2009b). However, as the climate has changed rapidly in recent decades (McGowan et al. 2018;Nicholls 2005), there have been significant changes in snowpatch plant communities in the Australian Alps over just 13 years.
These changes include temporary declines in diversity that stabilised in the latest survey, but responses varied across snowmelt gradients. Conversely, increasing diversity has been recorded in other long-term snowpatch monitoring sites Sandvik and Odland 2014). The earlier assessment of the seven snowpatches in the Australian Alps postulated that species richness would increase in the mid and late melt zones due to competitive invasion by more generalist species (Pickering et al. 2014), however, it appears from the latest survey that species richness has remained stable with few appearances or disappearances through time. Similarly, diversity within snowpatches in Japan has remained stable through time (Amagai et al. 2018). Here we reported few changes in diversity, which may be a consequence of nivation processes that have resulted in skeletal soils that hold little moisture and may restrict new species establishing in the centres of these snowpatches irrespective of lengthening growing seasons and increasing temperatures.
While we found dynamics in the cover of plants, growth forms and snowpatch specialists, there were varying responses across snowmelt gradients. In the early melt zone where plant cover was ~ 98% in 2007, increasing cover may be explained by the in-filling and expansion of larger and more competitive species resulting in densification, which has been recorded on nearby alpine summits (Verrall et al. 2021). We found increases in graminoid cover in the mid and late melt zones seems to be at the expense of snowpatch specialists, which are mostly forbs (Costin et al. 2000). Thus, there is cause for concern as current plant composition in mid and late melt zones deviates from the discrete plant communities dominated by dwarf forbs previously characteristic of these locations (Costin 1954;Pickering et al. 2014). Similar patterns of increasing plant cover over time has occurred in snowpatches in Japan (Amagai et al. 2018), Italy  and Switzerland (Matteodo Fig. 5 Principal components analysis of a species, b growth form, and c species association composition for each melt zone in 2007, 2013 and 2020. Asterisks indicate significant changes among years (*p < 0.05; **p < 0.01; ***p < 0.001). Black arrows represent successional trajectories of group means over time. The length and direction of biplot vectors demonstrate the relative forcing of influential com-ponents, quantified via SIMPER analysis of species (1a-Poa costiniana, 26.0%); 2a-Polytrichum juniperinum, 11.6%; 3a-Neopaxia australasica, 10.1%; 4a-Rytidosperma nudiflorum, 5.2%; 5a-Carex hypandra, 3.5%), growth forms (1b-graminoids, 48.4%; 2b-forbs, 26.7%, 3b-cryptogams, 24.9%) and snowpatch associates (1c-generalists, 62.6%; 2c-snowpatch specialists, 37.4%) Fig. 6 Dynamics in community-weighted trait means a plant height, b leaf area, c leaf fresh weight, and d leaf dry weight for the study area and for each melt zone for 2007, 2013 and 2020, with data points from each quadrat (colour-coded dots), distribution and density of data (colour-coded polygons), mean (black dot) and 95% confidence intervals (black line) shown. Asterisks indicate significant changes between years (*p < 0.05; **p < 0.01; ***p < 0.001) Fig. 7 Dynamics in the plant strategies per quadrat for the study area and for each melt zone for 2007, 2013 and 2020 with maximum extent convex hulls. Community-weighted strategies were calculated for quadrats by scaling the proportional abundance (cover) of each species competition-stress-ruderal proportion, as deduced from functional traits of each species (Pierce et al. 2017). Asterisks indicate significant changes between years (*p < 0.05; **p < 0.01; ***p < 0.001) et al. 2016). As the climate changes, dynamics in cover are seemingly occurring via competitive exclusion of snowpatch specialists by generalists.
We found that the greatest changes in composition over time took place in the mid melt and late melt zones, with clear trajectories towards the early melt zone. This suggests that compositional change may be amplified by the degree of microclimatic change, where plant communities that experience the shortest growing seasons are subject the greatest change (Good et al. 2019;Schöb et al. 2009). Here we show dynamics in species composition of snowpatches were primarily influenced by increases in tall tussock-forming graminoid Poa costiniana. Similarly, this was highlighted in the last assessment of these snowpatches (Pickering et al. 2014), and in other plant communities in the same alpine area reflecting broader patterns of change as conditions warm (Verrall 2018;Verrall et al. 2021). We also found that there was a subsequent decline in snowpatch specialists such as the dominant forb Neopaxia australasica, influencing changes in growth form and species association composition. Such changes provide further evidence for increasing competitive exclusion in many snowpatches, with successional encroachment from adjacent plant species and communities Matteodo et al. 2016;Schöb et al. 2009).

Dynamics in functional traits and plant strategies
Functional traits and plant strategies can help explain changes in alpine plant communities including along snowmelt gradients (Good et al. 2019). Here we show that the composition was increasingly dominated by taller plants with larger and heavier leaves over time, particularly in the mid and late melt zones. As these traits are linked to competitive light capture and drought tolerance (Cornelissen et al. 2003;Yin et al. 2019), the changes are consistent with abiotic conditions favouring species with different ecological strategies to those in the past. Similar to dynamics in cover and composition, these changes in functional traits reflect increases in generalist species such as tall tussock-forming graminoids (Pickering et al. 2014).
The changes recorded here also relate to plant strategies where species are classified on a continuum among three strategies: competitor (C), stress-tolerant (S) and ruderaltolerant (R) based on leaf area, leaf dry matter content and specific leaf area, respectively (Grime 2006;Pierce et al. 2017). At the start of monitoring in 2007, clear differences were apparent in plant strategies with the early melt zone dominated by stress strategists while the late melt zone was dominated by ruderal strategists. Similar patterns have been seen in snowpatches with longer growing seasons found at lower elevations in the Australian Alps (Good et al. 2019). However, we found that community-weighted strategies shifted from ruderal-tolerant to stress-tolerant in the late melt zone where snow cover is declining but temperatures and growing seasons are increasing through time. Similarly, there was also a shift towards stress-tolerant compositions in the mid melt and early melt zone.
Ruderal strategists may have previously dominated the areas with persistent snow cover as relatively short growing seasons require plants to develop quickly, in addition to the decline in periglacial and nivation disturbance regimes such as gelifluction, frost creep and snowpatch abrasion (Costin 1954;Costin et al. 1973Costin et al. , 1964. The shift from ruderal to stress strategist may reflect those changes as well as plants potentially experiencing more stressful conditions with declines in insulation provided by snowpatches in spring and autumn, and subsequent increases in exposure to extreme climatic conditions with greater fluctuations in diurnal temperatures and more frosts (Winkler et al. 2018;Wipf et al. 2009). Unexpectedly, the tall tussock-forming graminoids driving much of the changes were perceived as thermophilic competitors (Pickering et al. 2014;Verrall et al. 2021) while snowpatch specialists are generally considered to be stresstolerant (Good et al. 2019). However, the methods used to allocate plant strategies used here does not account for plant height (Pierce et al. 2017), and therefore does not capture any increasing competitive influence from taller graminoids in the snowpatches.

Microclimatic dynamics and implications for snowpatch processes
The microclimate of the late melt zone has changed significantly over the past 17 years, with growing seasons increasing by 36.7 days, with growing degree days and growing season mean temperatures increasing by 431.8°d and 1.49 °C, respectively. The mid melt zone was more variable, but growing seasons length and temperatures appear to be increasing, while there was little change in the early melt zone. However, despite these changes in microclimate over time, microclimatic differences are still apparent among the snowmelt zones. These results provide novel insights about how macroclimatic change relates to microclimatic change, which is particularly important as microclimatic heterogeneity of alpine ecosystems has been postulated to provide climatic refugia that may buffer the impacts climate change for alpine plants (Scherrer and Körner 2011). However, we show that the most pronounced changes occurred within the microclimatic refugia of the late melt zone, and these were strongly reflected by changes in the vegetation. Congruently, these snowmelt-dependant vegetation dynamics in response to climate change have been predicted and increasingly documented for snowpatches in Australia (Good et al. 2019;Pickering et al. 2014;Venn et al. 2011) and Europe 1 3 (Carbognani et al. 2012;Komac et al. 2015;Schöb et al. 2009). The temporal shift away from compositions of ruderal strategists towards taller, more competitive and stress strategists infers that the relative importance of key periglacial and nivation processes that historically maintained these distinctive plant communities may be declining. Therefore, it appears that these snowpatches in the Australian Alps may be approaching ecosystem collapse, with substantial loss of defining biotic and abiotic characteristics signalling replacement by a novel ecosystem (Bergstrom et al. 2021;Williams et al. 2015).
However, uncertainty remains about the future response of snowpatch plant communities to increasing temperatures and growing seasons, and will likely depend on a range of factors. This includes the plasticity of alpine plant species, but seems to be limited in Australia under rapid climate change (Briceno et al. 2014;Notarnicola et al. 2021;Sritharan et al. 2021), although snowpatch specialists are yet to be investigated. Furthermore, insights into the persistence of snowpatch plant communities have been gained in Europe by assessing seedbank plasticity and regeneration (Bernareggi et al. 2016(Bernareggi et al. , 2015Tonin et al. 2019), but such research is limited in Australia (Venn et al. 2011). Uncertainty also arises from the waning influences of abiotic factors in lieu of biotic interactions in alpine ecosystems as the climate warms (Klanderud et al. 2015;Margreiter et al. 2021;Schöb et al. 2012). However, the influence of biotic interactions has received limited attention in Australia so far (Jarrad et al. 2012). Another factor that may become important in Australian as snow declines is increasing trampling and grazing pressure from feral ungulates, as some snowpatch specialists had been almost grazed out of existence under historic pastoral land-use (Costin 1954). Therefore, investigations into the plasticity of snowpatch specialists, seedbanks and changing biotic interactions may provide valuable insights to limit the uncertainty of ecosystem collapse in snowpatches.
Nevertheless, there is increasing agreement that snowpatch plant communities in Australia will decline as the climate continues to warm (Camac et al. 2021), and now increasing evidence that this is occurring during this 13 year monitoring period of the seven snowpatches. While there are global long-term monitoring networks to monitor alpine summits (Pauli et al. 2015) and simulated environmental changes (Molau and Mølgaard 1996), no such initiative exists for periglacial alpine plant communities such as snowpatches (Verrall and Pickering 2020), despite amplified vulnerability to climate change (Gritsch et al. 2016;Matteodo et al. 2016). Therefore, it is important that the current snowpatch monitoring sites in the Australian Alps are maintained, in addition to the establishment of global monitoring networks for snowpatch plant communities as they may be among the first that are lost to climate change (Bergstrom et al. 2021;Williams et al. 2015).

Conclusion
This study provides the first long-term monitoring across snowmelt gradients for snowpatch plant communities, with changes in microclimate and vegetation most pronounced in areas where snow persisted the longest. With the decline of snowpatch specialists in the face of increases in temperatures and growing seasons, the erosion of key abiotic and biotic characteristics indicate snowpatches in the Australian Alps are approaching ecosystem collapse. However, other factors that may buffer against collapse are not accounted for in this study, such as the plasticity of snowpatch specialists, seedbanks and changing biotic interactions. Therefore, more research is required to better understand the fate of these rare plant communities in the face of climate change.