Introduction

Environmental changes are likely to cause large vegetation shifts in many polar and alpine regions (Rowland et al. 2016). Global warming may pose a serious threat to isolated endemic alpine species when neither upward nor poleward distribution shifts are possible (Kidane et al. 2019). In addition, global warming may seriously shrink suitable habitats (Ferrarini et al. 2019a) and cause local extinction of species at the extreme of their distribution range (Ferrarini et al. 2016; Hampe and Petit 2005). Global warming has already been shown to cause range shifts (Chen et al. 2011; Kullman 2002; Steinbauer et al. 2019) and composition shifts (Evangelista et al. 2016; Koltz et al. 2018; Liberati et al. 2019; Rosbakh et al. 2014). A widespread trend at many alpine and Arctic sites is an increase in shrubs (Jägerbrand et al. 2009; Maliniemi et al. 2018; Myers-Smith et al. 2011; Vowles and Björk 2019). Other plant groups (forbs and graminoids) have more inconsistent responses (Elmendorf et al. 2012; Vowles et al. 2017). However, a meta-analysis of 61 warming experiments in tundra vegetation found that graminoids tended to increase at colder sites, while showing neutral or negative responses at warmer sites (Elmendorf et al. 2012). Cushion plants, such as the widespread Silene acaulis, are predicted to be vulnerable to climate change in alpine areas (Doak and Morris 2010; Ferrarini et al. 2019a, b). However, many plant species in polar and alpine regions are long-lived (Ferrarini et al. 2019a; Morris and Doak 1998). Thus, even when conditions at a site become unfavourable due to environmental change, long-lived plant species can be expected to persist for more years than animals and short-lived species, as the plants are confined to the site and cannot migrate once established. In addition, longer lived plant species have been suggested to be less vulnerable to increased climate variability than short-lived species (Morris et al. 2008). However, we are not aware of any long-term (> 10 years) experimental studies that have tried to mimic the predicted increase of interannual variability of climatic factors, or different types of warming scenarios (ex. different levels of warming, increased warming with years, or occasional heat waves). Those conducted to date are short term. One previous study has applied a high level of warming, simulating a heat wave for 2 weeks in Greenland (Marchand et al. 2006), one study has applied low level, high level and increasing warming for 4 years in sub-arctic alpine Sweden (Alatalo et al. 2016), and one study has applied low and high levels of warming for 5 years in sub-arctic alpine Sweden (Jonasson et al. 1999).

Solar radiation and temperature have been shown to be dominant factors controlling net primary production in alpine meadows and grasslands in Tibet (Wang et al. 2018; Zheng et al. 2020) and Scotland (Gimona et al. 2006), while summer precipitation is an important driver for species richness in Tibet (Li et al. 2020). Extreme warming events that are accompanied by drought have been shown to have more severe effects on plant communities than warming without accompanying drought (Bragazza, 2008; De Boeck et al. 2016). From the growing number of experimental global change experiments performed, we have learnt that short- (1–3 years), medium- (4–7 years) and longer term (> 10 years) responses may differ (Alatalo et al. 2015b; Alatalo and Little 2014; Baruah et al. 2018; Hollister et al. 2005; Kremers et al. 2015; Walker et al. 2020).

Climate change experiments have been conducted at Latnjajaure field station in northern Sweden since 1993. An increasing number of studies at the field station are now covering potential impacts of climate change on different organism groups and ecosystem properties. In an experiment established at the station in 1995, on a nutrient- and species-poor heath and a more species- and nutrient-rich mesic meadow, we have previously reported on short- and medium-term responses and the impact of long-term warming on lichens (Alatalo et al. 2017a), bryophytes (Alatalo et al. 2020), soil fauna (Alatalo et al. 2015a, 2017b), and plant traits (Baruah et al. 2017).

In the present study, we examined the effect of 18 years of experimental warming on vascular plant communities. Specifically, we determined the impact on species composition, species diversity (at the level of rare, common and dominant species of each community), and phylogenetic and functional diversity of vascular plants. We hypothesised that (1) the effect of experimental warming is greater over time compared with in the early years of the experiment (Komatsu et al. 2019). Specifically, we hypothesised that long-term warming would have (2) a negative impact on species richness and diversity in the meadow, with well-developed vegetation cover, due to increased competition between species following lower environmental stress (as temperature increases), and (3) a positive impact on species richness and diversity in the heath, with its less developed vegetation cover allowing species to colonise new space.

Materials and methods

Study area

The study was conducted at Latnjajaure field station, which is located in the Latnjavagge valley (68°21´N, 18°29´E; 1000 m a.s.l.) in northern Sweden. The climate at the site is classified as subarctic (Polunin 1951), with snow cover for around 8 months of the year (usually from October to May), cool summers and relatively mild, snow-rich winters. The growing season starts in late May and ends in early September (Molau et al. 2005). Climate data were collected throughout the year at the weather station at Latnjajaure field station, with hourly means, maxima and minima recorded. Mean annual air temperature in the study period (1993–2013) ranged from  – 0.76 to  – 2.92 °C (Alatalo et al. 2017a). Mean monthly temperature was highest in July, ranging from 5.9 °C in 1995 to 13.1 °C in 2013 (Alatalo et al. 2017a). Mean annual precipitation during the period was 846 mm, but in individual years, it ranged from a low of 607 mm (1996) to a high of 1091 mm (2003) (Alatalo et al. 2017a). Detailed monthly mean, max and min temperature data and precipitation data (Alatalo et al. 2017a) are supplied in electronic supplementary materials to this paper. Physical conditions in the soils in the valley vary from dry, acidic and nutrient-poor to wet, base-rich and more nutrient-rich, with an associated variation in plant communities (Alatalo et al. 2017b; Björk et al. 2007; Lindblad et al. 2006; Sarneel et al. 2020).

The mesic meadow community has a vegetation averaging 67% canopy cover (Alatalo et al. 2017a), dominated by Carex vaginata, Carex bigelowii, Festuca ovina, Salix reticulata, Salix polaris, Cassiope tetragona, Bistorta vivipara and Thalictrum alpinum (Alatalo et al. 2014; Molau and Alatalo 1998). The more sparsely vegetated poor heath community (average 54% canopy cover) (Alatalo et al. 2017a) is dominated by Betula nana, Salix herbacea and Calamagrostis lapponica (Alatalo et al. 2015c; Molau and Alatalo 1998). The mesic meadow has thicker soil cover and higher soil nutrient and moisture content than the heath (for details, see Alatalo et al. (2017b)).

Experimental design and measurements

In July 1995, twelve 1 m × 1 m plots with homogeneous vegetation cover were marked out in an alpine mesic meadow plant community and in a heath plant community, and randomly assigned to treatments (control and experimental warming) in a factorial design. At the start of the experiment, there were eight control plots and four plots with experimental warming (total 12) in each plant community. However, as we could not identify all initial control plots in 2013, we only made measurements in four control and four experimental warming plots in each community in 2013. The experimental site was dismantled in 2016 (2013 being the last year of plant measurements). For 1995–2001, we used data from eight control plots. Experimental warming is applied at the site using open-top chambers (OTCs) left on plots with warming treatment all year around. In the initial years, the temperature in the control and OTC plots was monitored with Delta™ and Tinytag™ loggers (Molau and Alatalo 1998). As found in other studies (Hollister and Webber 2000; Marion et al. 1997; Molau and Alatalo 1998), the OTCs increased the air temperature by 1.5–3 °C compared with control plots with ambient temperature (Molau and Alatalo 1998). OTCs have also been shown to decrease canopy moisture (Hollister and Webber 2000), causing earlier snow melt and prolonging the growing season (Hollister and Webber 2000; Molau and Alatalo 1998).

Abundance of all species was assessed using a 1 m × 1 m frame with 100 grid points (Walker 1996) in the middle of the growing season in 1995, 1999, 2001 and 2013. Due to their hexagonal shape, the OTCs reduced the number of points per plot to 77–87, and thus warmed plots had fewer pin-point intercepts than control plots. To compensate for this, we analysed the relative changes from 1995 as suggested by Kent et al. (2011). Fixed points at the corner of each plot allowed the grid frame to be placed in the same position on the plot on each measuring occasion. This method has been shown to be accurate in detecting changes in tundra vegetation (May and Hollister 2012).

Data analysis

Community composition

The effect of the warming treatment over time on species composition in both plant communities was evaluated using principal response curves (PRC), which show the community-level treatment effects over time compared with the control and enable species-level changes to be interpreted (Van den Brink and Braak 1999). Monte Carlo permutation tests were performed to evaluate the statistical significance of differences between each treatment and the control. The PRC and permutation tests were performed using the vegan package (Oksanen et al. 2017).

We also evaluated temporal variation in species composition within each plant community in each year (i.e., within-site beta diversity), to test whether small-scale (i.e., plot) conditions can lead to different responses in the plant communities in the area. Hellinger distance (i.e., Euclidean distance of the Hellinger-transformed data) was used as a measure of within-site beta diversity. This dissimilarity index was calculated using the vegdist function in the vegan package. For each year, mean and 95% confidence intervals (CIs) were calculated. The CIs were estimated using a one-mean t-procedure (Zar 2010). All of the calculated indices were relativised, using within-site beta diversity in 1995 as the base value. R version 3.5.3 was used for the analyses (R Core Team 2019).

Species diversity

Hill species diversity indices were calculated to compare changes in the species diversity of the heath and meadow communities between the sampling years. These indices are considered the standard tool for calculating and comparing species diversity (Erfanian et al. 2019a). We considered species richness (q = 0 in the Hill species diversity formula), the exponential of Shannon diversity (q = 1) and the reciprocal of the Simpson index (q = 2), which reveal the diversity of rare, common and dominant species, respectively (Chao et al. 2014a, b; Erfanian et al. 2019b). Sampling in the present study was conducted during several years. Unequal sampling effort between sampling years, which greatly affects biodiversity estimates, is a typical limitation of this type of study (Kent 2011). To eliminate the effects of this limitation on our inferences, we used a coverage-based rarefaction/extrapolation method, where the species diversities in the different years were calculated at the same coverage (i.e., sampling effort) level (Chao et al. 2014b; Chao and Jost 2012). The 95% CIs for the estimated diversities were calculated, using a bootstrapping approach. These analyses were performed in the iNEXT package, using the estimateD function (Hsieh et al. 2016). All of the calculated indices were relativised, using the species diversity in 1995 as the base value. Positive and negative values of relativised diversity indicate higher or lower diversity, respectively, compared with the 1995 baseline.

Phylogenetic diversity

The phylogenetic diversity of the communities was used, since it reflects the evolutionary history of the assemblages and is related to their conservation value (Faith 2016; Faith and Baker 2006). The phylogenetic tree of the vascular plants collected from plots was estimated using the V. PhyloMaker package (Jin and Qian 2019). Hill diversity indices of phylogenetic diversity at the level of rare (q = 0), common (q = 1) and dominant (q = 2) species were considered. The coverage-based rarefaction/extrapolation method was employed to calculate these indices at the same coverage level. The iNEXT-PD package was used for the calculations (Chao et al. 2010; Hsieh et al. 2016). The results obtained were relativised, using the phylogenetic diversity in 1995 as the base value.

Functional diversity

Changes in the dominance structure of five functional groups of vascular plants (cushion-forming plants, deciduous shrubs, evergreen shrubs, forbs and grasses) were evaluated. Cushion plants included Diapensia lapponica, Harrimanella hypnoides, Saxifraga oppositifolia and Silene acaulis. All these species can form more or less dense cushions depending on the physical conditions at the growing site. For example, the growth form of S. acaulis can change substantially along an elevation gradient (Bonanomi et al. 2015). Due to low numbers of observations in some groups, we could not statistically compare these results. Boxplots were drawn to visualise relative changes in functional group abundance between control and warming plots. For relativisation, abundance data for control and warming plots were subtracted from the median of the abundance of control plots in 1995. The resulting value was divided by the median abundance in control plots in 1995. This result was multiplied by 100, to show the percentage change in functional group abundance during the 18 years of the experiment. Thus, negative values in boxplots indicate decreased abundance, while positive values show increased abundance of functional types compared with the median of the control community. For visualisation, the percentage values were log-transformed. The Hellinger distance was calculated separately for each functional group, to assess the species turnover within groups.

Results

Species composition

The PRC analysis results revealed a significant difference (p value = 0.039, F value = 7.166) between control and warming plots of the heath community in terms of species composition. The PRC analysis explained 16% of the variance, and 15.35% of this variance contributed to the first axis (Fig. 1). For the meadow vegetation, the PRC analysis did not detect a significant difference (p value = 0.659, F value = 1.814) between species composition of control and warming plots. About 9% of variance was explained by the PRC analysis and 7.56% of this explained variance contributed to the first axis. Only species with relative frequency sum above 1 are shown in Fig. 1. In meadow vegetation, Carex vaginata showed the greatest average abundance increase and Cassiope tetragona showed the greatest average abundance decrease (Fig. 1). In heath vegetation, Betula nana showed the greatest increase in average abundance and Empetrum hermaphroditum showed the greatest decrease (Fig. 1).

Fig. 1
figure 1

Principal response curve (PRC) showing the effect of warming treatment over time on vascular plant species in heath and meadow vegetation at Latnjajaure, northern Sweden. Only species with relative frequency sum > 1 are shown

Full size image

Long-term warming led to a significant decrease in beta diversity in the heath plots over time (Fig. 2), but to increased beta diversity in the meadow plots (Fig. 2). From 2001 to 2013, the beta diversity in warming plots stabilised for both vegetation types. In 2013, for both vegetation types, there was no significant difference between beta diversity of control and warming plots. In the heath, the difference between control and warming plots was significant from 1999 until 2001, while in 2013 the confidence intervals overlapped slightly. In the meadow, the difference between control and warming plots was significant in 1999 and 2001. In meadow control plots, beta diversity started to increase from 2001.

Fig. 2
figure 2

Relative changes in within-site beta diversity in response to long-term warming (1995–2013) in an alpine heath community and a meadow community at Latnjajaure, subarctic Sweden. Values represent mean and 95% confidence intervals. 1995 data used as baseline

Full size image

Species diversity

For meadow vegetation, except in the year 1999, warming had a significant negative effect on all three orders of Hill species diversity (i.e., q = 0, 1 and 2) compared with the control (Fig. 3).

Fig. 3
figure 3

Changes in species diversity in response to long-term warming (1995–2013) in an alpine heath community and a meadow community at Latnjajaure, subarctic Sweden. Species diversity is shown at the level of rare, common and dominant species in the community, indicated by q = 0 (species richness), q = 1 (exponential of Shannon diversity) and q = 2 (reciprocal of Simpson index), respectively. 1995 data used as baseline

Full size image

For heath vegetation, except for dominant species in 2001, there were no significant differences in species diversity (i.e., species richness (q = 0, including rare species), the exponential of Shannon diversity (q = 1, or common species) and the reciprocal of the Simpson index (q = 2, or dominant species)) between warming and control plots (Fig. 3).

Phylogenetic diversity

Comparison of phylogenetic diversity (hereafter PD) estimates for control and warming plots in the meadow showed that there was a significant difference between these two treatments in the 2001 inventory (Fig. 4). In 1999, PD at the level of q = 1 (common species) and 2 (dominant species) differed significantly between the control and warming treatment. In 2013, only PD at the level of q = 1 (common species) showed a significant difference between control and warming plots. In all the above cases, warming had a negative effect.

Fig. 4
figure 4

Changes in phylogenetic diversity (Hill diversity indices at the level of rare (q = 0), common (q = 1), and dominant (q = 2) species) in response to long-term warming (1995–2013) in an alpine heath community and a meadow community at Latnjajaure, subarctic Sweden. Values represent mean and 95% confidence intervals. 1995 data used as baseline

Full size image

In the heath vegetation, at the level of q = 2 warming plots showed significantly lower PD than control plots (Fig. 4). This was also observed in 2013 at the level of q = 1. No significant difference was detected between control and warming plots at the PD level of q = 0.

Functional diversity

In meadow warming plots, abundance of cushion plants remained unaffected over time, but abundance of deciduous shrubs increased (Fig. 5). Control plots of both vegetation types also showed increased abundance of deciduous shrubs from 1995 until 2001. However, by 2013, the abundance of deciduous shrubs had decreased in the control plots in both communities (Fig. 5). In the meadow, evergreen shrub abundance remained similar over time in control plots, but increased in warming plots. In the heath, abundance of evergreen shrubs increased over time in both warming and control plots (Fig. 5). Forb abundance remained unchanged in the warming plots in the meadow, while it decreased over time in the warming plots in the heath (Fig. 5). Graminoid abundance remained stable over time in both control and warming plots in the meadow, and in warming plots in the heath. However, graminoid abundance decreased in the control plots in 2013 (Fig. 5).

Fig. 5
figure 5

Boxplots showing relative changes in abundance of cushion plants, deciduous shrubs, evergreen shrubs, forbs and graminoids over time in control (green) and warmed (red) plots in meadow (top panel) and heath (bottom panel) vegetation at Latnjajaure, northern Sweden

Full size image

Changes in the species composition of each functional group, measured using the Hellinger dissimilarity measure, are presented in Table 1. Cushion-forming plants, deciduous shrubs and evergreen shrubs showed low species turnover in both warmed and control plots. In contrast, forbs and graminoids showed moderate species turnover from 1995 to 2013 in both warmed and control plots.

Table 1 Changes in species composition (measured as Hellinger dissimilarity) of different plant functional groups in response to long-term warming (1995–2013) in an alpine heath community and a meadow community at Latnjajaure, subarctic Sweden
Full size table

Discussion

This study examined the impact of long-term experimental warming on species composition and species, phylogenetic and functional diversity. Our hypothesis (1) was partially supported, since the effect of experimental warming was greater from year seven of the experiment in the meadow, but not in the heath. Moreover, in the meadow, the 7-year responses were more pronounced than the responses in year 18. Our hypothesis (2) was fully supported, as long-term warming had a negative impact on species richness and diversity in the meadow. However, long-term warming had no positive effect on species richness and diversity in the heath, contradicting our hypothesis (3).

Ambient temperature at the experimental site showed an increase of around 2 °C during the study period (Alatalo et al. 2017a). Thus, plants growing in control plots also experienced warmer conditions and plants in the warmed plots were exposed to a greater temperature increase than initially intended (~ 2 °C experimental warming plus ~ 2 °C ambient warming). Long-term warming (18 years) drove differentiation in species composition of the heath vegetation over time, with the warmed plots ending up with distinctly different species composition in 2013 compared with 1995. In the meadow, warming caused deciduous and evergreen shrubs to increase in abundance, while graminoids as a group remained stable. However, warming caused some graminoid species (e.g. Carex vaginata and Festuca ovina) to increase in abundance in the meadow. A previous study at the same site found that 7 years of experimental warming caused sedges to decline in the meadow (Alatalo et al. 2014). Thus, the short-term and longer term responses differed. Previous studies have found that the effects of experimental treatments increase with time (larger effects after 10 years) (Komatsu et al. 2019). Similarly, we found that the majority of changes occurred from year 7 in our study period. Many previous studies have reported increased occurrence of shrubs in alpine and arctic tundra ecosystems, and have attributed this to ongoing climate change (Jägerbrand et al. 2009; Maliniemi et al. 2018; Myers-Smith et al. 2011; Myers-Smith and Hik 2018; Vowles and Björk 2019). Our results show that the responses can vary considerably even on local scale, as warming increased deciduous shrubs markedly in the meadow plots, but not in the nearby heath plots studied at the site, while evergreen shrubs increased in both communities. Deciduous shrubs showed a similar positive effect of experimental warming in the initial 5-year response at our site (Jägerbrand et al. 2009), and in Alaskan Tundra (Chapin III and Shaver 1985, 1996). In addition, a long-term monitoring study in High Arctic Canada experiencing natural warming found that evergreen shrubs, but not deciduous shrubs, increased over a period of 27 years (Hudson and Henry 2009). Changes in species composition have also been reported for grasslands in Tibet (Liu et al. 2018), Oklahoma (Shi et al. 2018) and the Pyrenees (Boutin et al. 2017), for snowbed and nival vegetation in the European Alps (Lamprecht et al. 2018; Matteodo et al. 2016) and for tussock tundra in Alaska (Leffler et al. 2016).

Variability in the species composition of plots increased in the meadow vegetation, while it decreased in the heath vegetation. This finding suggests that the meadow vegetation responded to climate changes in different ways. Heath vegetation showed a poor adaptive response, and we observed compositionally homogenised communities. This is a negative change, as homogenised communities can potentially be more vulnerable to future disturbances. Considering the differing responses of heath and meadow communities, we concluded that the heath vegetation was more susceptible to climate change impacts. A previous study in the Swiss Alps revisiting 63 sites experiencing natural warming over time found that arrival of new species resulted in homogenisation of the plant communities (Matteodo et al. 2016). However, the stability of species composition varied between plant communities, with snowbed communities being more vulnerable than grassland communities (Matteodo et al. 2016).

While not applying experimental warming, a monitoring study over 40 years in alpine Colorado found that species richness declined in all three plant communities studied (dry meadow, moist meadow and shrub tundra), with the largest decline in the shrub community (Scharnagl et al. 2019). The two plant communities at our study site in northern Sweden responded with contrasting patterns at different levels of species diversity (rare, common and dominant species) to the ambient temperature and experimental warming treatments over time. Experimental warming caused an initial negative response in within-site diversity in the heath (Alatalo et al. 2015c), which remained negative in the long term (this study). While the heath experienced a decrease in dominant species (q = 2) evergreen shrubs increased in abundance. In alpine Colorado, increases in shrub abundance and plant litter under ambient climate conditions was accompanied by a decrease in forbs, graminoids and non-vascular plants (Scharnagl et al. 2019). The meadow community at our study site showed an initial rapid negative response (until 2001), after which it started to recover, but it had not returned to its initial status after 18 years of warming. In terms of phylogenetic diversity, the long-term warming had a significant negative effect on the three orders of phylogenetic Hill diversity in the meadow. While there was a similar tendency in the heath, only phylogenetic diversity of dominant species was significantly affected. The observed reduction in phylogenetic diversity of both communities can be considered an indication of loss of rare and phylogenetically diverse species, as the final colonising species have lower phylogenetic diversity, because they come from related taxa. Notably, forbs and graminoids in meadow showed larger turnover in species composition during the 18-year experiment, while shrubs showed much lower turnover in the heath. However, these differences in species turnover could potentially be explained by differences in their longevity, with forbs and graminoids including more short-lived species than shrubs.

In a previous study at our site, it was shown that 7 years of warming caused a significant decline in total species richness (Alatalo et al. 2014). These results support indications in other studies that mesic meadow communities tend to be more vulnerable than drier sites in terms of species loss (Elmendorf et al. 2012). However, several studies have shown that plants in tundra and alpine ecosystems are often nutrient-limited (Eskelinen et al. 2017; Haag 1974; Soudzilovskaia et al. 2005). As nitrogen mineralisation has been shown to increase due to increased microbial activity in response to warming, plants on thicker, more nutrient-rich soils that are normally limited by temperature/short growing season can be expected to respond more positively to warming/extended growing season than plants growing in thinner soils with nutrient-poor conditions (Rustad et al. 2001). However, microbial responses to warming in alpine areas also depend on precipitation levels (Hu et al. 2020). In our long-term experiment, long-term warming caused a decrease in soil moisture in the meadow community, but not in the heath community (Alatalo et al. 2017b). This could potentially help to explain the differences in responses between these plant communities. Decreased soil moisture due to experimental warming has been reported to be accompanied by a decrease in sedges and an increase in grasses and forbs in a meadow community in Tibet following short-term experimental warming (Peng et al. 2017), with meadows becoming more similar to heaths in alpine Sweden (Scharn et al. 2021). However, at our meadow site we found a more complex response pattern over time, as sedges decreased initially (Alatalo et al. 2014), but increased over the longer term (this study). This later increase in sedges was mainly driven by Carex vaginata, while the initial shorter term responses were dominated by changes in abundance of Carex bigelowii (Jägerbrand et al. 2009). Thus, negative effects on species and phylogenetic diversity may be driven indirectly by decreased moisture levels due to warming, not by the warming itself (Scharn et al. 2021). In addition, both the responses and major drivers (species) may change over time.

Cushion-forming plants are important in alpine areas due to their function as facilitator species (Anthelme et al. 2014; Cavieres et al. 2014). The shorter term results (1995–2001) from our experiment showed that the dominant cushion-forming plant at the site, Silene acaulis, was highly plastic in its phenotypic responses in terms of growth-related plant traits to nutrient addition and combined nutrient addition and warming, while warming alone had no effect on growth and abundance (Alatalo and Little 2014). In the present longer term study, the PRC showed that S. acaulis only decreased slightly in response to 18 years of warming, so the studied population is likely resistant to warming that is not accompanied by an increase in nutrients. The dominant cushion plant in the meadow, S. acaulis, has a taproot. As the soil became drier in the warmed plots (Alatalo et al. 2017b), having a taproot could have provided an advantage over more shallow-rooted species in the longer term. However, a previous study has shown that S. acaulis populations across the species distribution range may respond in different ways, with southern populations of S. acaulis having higher growth rates than northern populations in North America, but lower survival and recruitment (Doak and Morris2010). That study also found that the warmest years had a negative effect on survival and fruit production, but that moderately warmer years had a positive effect (Doak and Morris 2010). In contrast, a recent study showed that northern populations of S. acaulis may decline, while southern populations may remain stable (Peterson et al. 2018). This highlights the difficulty in predicting plant species responses to climate change, as both life history plasticity and local adaptation will affect species responses to warming (Peterson et al. 2018).

In addition, different types of environmental disturbance, such as grazing and land degradation status (depending on the status of human land use intensity) (Erfanian et al. 2019a), experimental warming, nutrient addition and changes in light regime (Jonasson et al. 1999), may have different effects on plant communities. Strong evidence of the importance of both duration of experimental manipulations and number of disturbances is provided by a global study that included data from more than 100 experiments (Komatsu et al. 2019). The study showed that the greater the number of experimental perturbations and the longer the experiment, the less resistant plant communities were to the experimental treatments (Komatsu et al. 2019). Plant communities were in general resistant to experiments that ran for less than 10 years, while experiments lasting more than 10 years showed larger changes. In addition, plant communities that were exposed to three or more experimental treatments showed larger changes in community composition than plant communities that experienced fewer environmental manipulations (Komatsu et al. 2019). Thus, short-term responses may be poor predictors of potential long-term changes. It is, therefore, important to try to ensure that global change experiments are maintained and re-sampled over longer periods than normally covered by external funding for research projects. Unfortunately, our experimental site was dismantled in 2016.

Conclusions

This study found that responses in plant species composition and phylogenetic diversity to experimental warming varied both in time (medium vs long term) and space (neighbouring heath and meadow communities). The heath community was more negatively affected in terms of species composition and community-level responses than the meadow community. However, the meadow community showed a larger decrease in species and in phylogenetic diversity than the heath community. Long-term warming caused differentiation in species composition in the two communities, with shrubification in both communities and decreases in forbs in the heath community. A potential driver for the changes in the meadow community may be decreased soil moisture caused by the long-term warming.