Recurrent summer drought affects biomass production and community composition independently of snowmelt manipulation in alpine grassland

Earlier snowmelt and more frequent summer drought due to climate warming are considered particularly influential for extratropical alpine plants, which are adapted to a short growing season and high water availability. Here, we explored the combined effects of the two drivers with a field experiment in late‐successional alpine grassland in the Swiss Alps (2500 m a.s.l.) over 6–7 years. We advanced and delayed snowmelt by removing and adding snow to experimental plots prior to natural snowmelt for 7 years and combined this treatment with 5 and 10 weeks of summer drought for 6 years. We measured plant biomass formation, community composition and ecosystem respiration, and monitored soil moisture as well as soil temperature. Natural snowmelt dates varied by 42 days across years. Snow manipulations advanced and delayed snowmelt by 4.6 and 8.0 days on average but did not affect annual growth (peak biomass) above‐ nor below‐ground. Interactions between snowmelt and drought were nonsignificant, implying that drought effects were independent of snowmelt. Drought reduced total annual above‐ground biomass in the 10‐week treatment by 16 ± 7% across years, while the 5‐week treatment lowered biomass in the last year only. This decline in biomass was accountable to high drought sensitivity of biomass production in a few forb and graminoid species. In contrast, drought did not affect the biomass production of the dominant sedge Carex curvula, whose proportion of total plant cover increased from 36% in controls to 48% in 10‐week drought. Below‐ground biomass slightly increased under drought (5‐week treatment only), resulting in a higher root mass fraction (both treatments). Despite continued root formation, drought reduced ecosystem respiration by 13%–23% per season, assessed nine times during three growing seasons. Since more than 85% of ecosystem respiration stemmed from below‐ground activities and roots continued growing under drought, we assume that soil microorganisms were heavily constrained by the drought treatments. Synthesis. We conclude that snowmelt timing is unrelated to productivity, while recurrent drought will shift biomass allocation from shoots to roots in this typical alpine grassland, with potential implications for grazers but also for nutrient and carbon cycling. Species‐specific drought‐sensitivity will considerably alter community composition under more frequent drought.


| INTRODUC TI ON
The effects of climate change are deemed especially pronounced in alpine regions, where plants are adapted to a cold environment (Körner, 2021) and temperature is rising above-average (Pepin et al., 2015(Pepin et al., , 2022)).Beside temperature, other, indirect consequences of a warmer atmosphere may play a decisive role for plant functioning at high elevation, such as earlier snowmelt (Stewart et al., 2004;Vorkauf, Marty, et al., 2021) and restricted water availability during summer (Calanca, 2007;Gobiet et al., 2014;Kotlarski et al., 2023).In the Swiss Alps, snowmelt was already occurring 17 days earlier in 2019 than in 1958 and is projected to further advance by up to 1 month until the end of the 21st century (Vorkauf, Marty, et al., 2021).This will significantly extend the time available for growth of alpine plants, which are adapted to a growing season length of only 2-3 months (Körner, 2021).At the same time, summer drought is becoming more frequent in the Alps: the probability of a drought event of the same magnitude as the severe drought in 2003 was 4% in the 20th century but could rise to 25% in 2071-2100 (Calanca, 2007).Thus, shifted snowmelt and more frequent drought could potentially have a greater impact on alpine plants than a few degrees higher air temperature, especially considering the favourable microclimate in which alpine plants grow (Körner, 2021;Scherrer & Körner, 2010).Yet, to our best knowledge, experimental studies that address the combined consequences of altered snowmelt timing and recurrent summer drought on alpine grasslands do not exist.
Earlier snowmelt and reduced water availability may alter productivity and dry matter allocation of alpine grassland species, potentially impacting nutrient cycling, species interactions, and community composition.Moreover, herbivores on managed (e.g.sheep and other livestock) and unmanaged lands (e.g.grasshoppers, marmots, snow hares, chamois and ibex) depend on the annual biomass production of these grasslands (Blumer & Diemer, 1996;Jäger et al., 2020).As biomass production may correlate with growing season length (Kudo, 1992), earlier snowmelt was expected to enhance productivity (Rammig et al., 2010).However, experimental studies have provided little evidence of a stimulating effect of earlier snowmelt on biomass production so far (Baptist et al., 2010;Boyle et al., 2022;Darrouzet-Nardi et al., 2019;Möhl et al., 2022;Wipf et al., 2009).On the other hand, drought periods are known to reduce above-ground biomass production in different grassland types (Kahmen et al., 2005;Kröel-Dulay et al., 2022;Mackie et al., 2019), including subalpine meadows (Gilgen & Buchmann, 2009;Ingrisch et al., 2018).This effect could exacerbate with earlier snowmelt in the future, because a smaller fraction of the snow-free period profits from the high soil moisture following snowmelt.Given the high interannual variability of snowmelt dates, identifying interactions with drought requires several years of in-situ experimental data.
Longer term experimental studies (>5 years) on summer drought or snowmelt timing in alpine grassland are scarce, although the longterm aspect is a prerequisite for distinguishing between transient and permanent responses to environmental change (Lindenmayer et al., 2012).Our knowledge of how drought affects productivity of alpine vegetation stems mainly from short-term studies (1-2 seasons): 1-9 weeks of rain exclusion per season resulted in higher root:shoot ratios by reducing above-ground biomass (de Boeck et al., 2015;Johnson et al., 2011), while below-ground biomass remained unchanged (Gilgen & Buchmann, 2009).An increased root:shoot ratio in response to reduced precipitation (−50%) was also observed in grassland of the Tibetan Plateau (Liu et al., 2018), although in this case root biomass increased and above-ground biomass remained stable.Drought also lowers microbial activity (Schimel, 2018), plant carbon allocation to microbes (Karlowsky et al., 2018), and root respiration (Hasibeder et al., 2015), with pronounced effects on ecosystem respiration.Two 8-days water exclusions within 1 month led to reduced above-ground biomass, net CO 2 exchange, and ecosystem respiration in monoliths of Nardus stricta dominated grassland (Johnson et al., 2011).Similarly, soil and ecosystem respiration of subalpine grasslands was up to 60% lower after 6-9 weeks of rain exclusion (Hagedorn & Joos, 2014;Ingrisch et al., 2018) but recovered quickly or even surpassed controls during the recovery period.
Most studies rely on daytime assessments of ecosystem respiration, despite the drawbacks associated with measuring respiration during the day, especially in alpine ecosystems with strong solar radiation.
Sudden darkening can lead to strong microclimatic changes (light, temperature, humidity) and trigger substantially increased respiration in plants ('light-enhanced dark respiration', Atkin et al., 1998;up growing under drought, we assume that soil microorganisms were heavily constrained by the drought treatments.6. Synthesis.We conclude that snowmelt timing is unrelated to productivity, while recurrent drought will shift biomass allocation from shoots to roots in this typical alpine grassland, with potential implications for grazers but also for nutrient and carbon cycling.Species-specific drought-sensitivity will considerably alter community composition under more frequent drought.

K E Y W O R D S
climate change, ecosystem respiration, high elevation, long-term, manipulative experiment, root mass fraction, root production, snowmelt to +30% in alpine grassland, T. Gross & E. Hiltbrunner, unpublished).
Hence, in situ measurements of dark respiration are most reliable when conducted at night.Alpine grasslands are often species-rich and the impacts of environmental changes are known to differ among grassland species (Suttle et al., 2007).Previous studies highlighted that mountain species belonging to different functional types (e.g.grasses and forbs) respond differently to climate change (Gilgen & Buchmann, 2009;Liu et al., 2018;Rosbakh et al., 2017).Plant species with high phenological flexibility may benefit from earlier snowmelt, while those species capable of forming dense root systems or symbiotic networks may gain an advantage during drought (Freschet et al., 2021;Jumpponen & Trappe, 1998).Such differential responses to snowmelt timing and drought may lead to future changes in community composition in alpine grassland (e.g.Liu et al., 2018).Despite that, species composition in late-successional subalpine and alpine grasslands was found to be stable in the face of the changing climate of recent decades with increasing air temperatures (Vittoz et al., 2009;Windmaißer & Reisch, 2013), but further environmental change and recurrent extremes may disrupt this stability.
We experimentally advanced and delayed the timing of snowmelt in a widespread alpine grassland type and combined these treatments with either 5 or 10 weeks of rain exclusion during the growing season.For six consecutive years (2017-2022), we measured above-ground biomass production at the species-level and assessed total root production using in-growth cores for two 2-year periods.In 2021, we additionally harvested standing root biomass to quantify net shifts in biomass allocation.We complemented these proxies for productivity with pointwise, night-time measurements of ecosystem respiration, which we conducted three times per season from 2019 to 2021.We hypothesise that (1) biomass declines aboveground and increases below-ground to compensate water shortage under both drought levels.(2) Night-time ecosystem respiration declines progressively with increasing drought duration.(3) Effects of snowmelt timing on biomass production materialize after several years only, while earlier snowmelt immediately amplifies adverse effects of summer drought.(4) We also expect that the response to snowmelt and drought differs between species and functional groups, and hence, leads to changes in community composition.

| Study site, vegetation and soil
Our experiment was located on a SW-oriented, slightly inclined (ca. 10°) slope at 2450 m a.s.l. in the Swiss Alps (46°33′47′′ N, 8°23′28′′ E; Figure S1; permission granted by the municipality of Obergoms).Vegetation is dominated by the sedge Carex curvula All.s.s.(referred to as Carex), which makes up around 30%-45% of annual biomass production above-ground (Möhl et al., 2020;Schäppi & Körner, 1996).This grassland community, termed Caricetum curvulae (Braun-Blanquet, 1948), is widespread in European mountain systems (Puşcaş & Choler, 2012) and the most frequent alpine grassland on acidic bedrock in the Swiss Alps (Landolt, 2012).At least 34 species grow at the site, 23 forbs and 11 graminoids.Beside the annual forb Euphrasia sp., all species are perennial, mostly of short stature and with below-ground shoot meristems.The soil is a ca.60-80 cm deep, acidic, partly podzolic cambisol with a soil pH CaCl2 between 3.6 (0-10 cm) and 4.2 (10-20 cm).The soil C:N ratio is 11.7 at 0-10 cm depth and increases to 13.9 at 10-20 cm depth with 1.8 and 1.0 mg N per g dry soil in the two soil horizons, respectively.Skeleton content increases along the soil profile from a few percent in the top 15 cm to 10%-25% at 30 cm and >40% at ca. 70 cm depth.Root density is extremely high in the uppermost 10 cm, but decreases exponentially with soil depth (only few roots below 50 cm), which is typical for alpine grassland (Körner, 2021).The permanent wilting point, derived from replicated pF curves, is 10.1 vol-% at 10 cm and 8.8 vol-% at 30 cm soil depth (Vorkauf, Kahmen, et al., 2021).Precipitation during alpine summer amounts to roughly 100-150 mm per month (June-September) and snow load during winter reaches 2-4 m.

| Experimental treatments
We defined 45 plots (2 m × 2.5 m) across five replicated blocks (Figure S1).(2019)(2020)(2021)(2022) to delay snowmelt (Figure S1).In 2018, heavy thunderstorms delayed snowmelt manipulations and the remaining snow height was only 0.5 m.Therefore, we reduced it to 0.3 m in early snowmelt plots, while the thin snow pack only allowed raising snow height to 0.7 m in late snowmelt plots.Additionally, we covered the piled up snow with white fleece to further delay snowmelt (Datex KN 25; Fritz Landolt AG).The fleece covers worked well and this practice was thus continued in the following years (2019-2022), allowing us to lower the target snow height for late snowmelt to 1.3 m and still achieve similar or even longer delays in snowmelt as in the initial 2 years.Fleece covers were removed once controls became snow free, which was 10-30 days after snow manipulations.
We employed two rain-exclusion treatments that started after snowmelt and lasted for either 5 or 10 weeks ('moderate' and 'extreme' drought, respectively), which is in the range of previous studies in subalpine grassland (7-9 weeks in Gilgen & Buchmann, 2009; 6 weeks in Ingrisch et al., 2018).Drought was realized with tent-like rain-out shelters that exceeded the plots on all sides (2.5 m × 3 m; ridge height of 1.2 m; Figure S1), consisting of aluminium frames with UV-B permeable, transparent plastic foil (Lumisol AF clear with ca.90% light transmission; Hortuna AG).From 2017 (1 year later than the first snow manipulations) to 2022, 15 plots were sheltered for 5 weeks per season, 15 plots for 10 weeks and the remaining 15 plots were never covered by rain-out shelters (controls).Shelters were installed immediately after snow had melted in all plots, which varied between years (Figure 1).In 2018, 2019 and 2022, naturally occurring drought periods forced us to water control plots on one to two occasions per season with 17 L m −2 per plot, which is enough to support normal evapotranspiration for 4-5 days (van den Bergh et al., 2013) and similar to an average rain event at this site (Figure 1).
In each plot, we marked a central area of 1 × 1 m (central-m 2 ) that excludes edge effects (Figure S1).

| Environmental conditions
Air temperature, radiation and precipitation were monitored 1.5 m above-ground every 10 min by two on-site weather stations during the growing season starting in 2017 (Vantage Pro2; Davis Instruments Corp.).In the central-m 2 of each plot, soil temperature was read once per hour at 3-4 cm depth (TidbiT v2 Temp; Onset Computer Corp.), which is close to most species' meristems.Soil temperatures were recorded year-round from 2016 to 2022 and allowed to accurately delineate snowmelt date in spring, defined as the first hour when mean temperature over 24 h rose above 3°C.Every autumn, T-sensors were read out and replaced if needed.In five cases (out of 315), T-sensors failed during winter and snowmelt timing was determined from webcam images or nearby plots with the same snowmelt treatment.We measured topsoil moisture content (0-5 cm) every 1-2 weeks in in each corner and in the centre of every plot during the growing season (Theta probe ML2 and ML3; Delta-T Devices).From 2018 onwards, soil moisture was simultaneously assessed at 5-10, 15-20, 25-30, and 35-40

| Above-ground biomass
We harvested peak biomass as a surrogate for annual biomass production above-ground, which is widely adopted in such grassland types where only a negligible amount of biomass is produced later in F I G U R E 1 Weather conditions at the experimental site in 2017-2022.Blue bars show daily sums of precipitation, whereas red and orange lines show mean air (1.5 m above-ground) and soil temperature (3-4 cm depth, mean of control plots), respectively, measured on site.Red dotted lines at the start and end of the season show air temperature from the nearby research station (2 m above-ground).Dotted black lines (2019)(2020)(2021)(2022) indicate topsoil moisture in control plots (vol-%, same y-axis as precipitation).Dark grey background indicates the sheltering period of the 5-week drought treatment and the entire grey background (dark + light grey) the duration of the 10-week treatment.Sums of excluded precipitation (Σ) were calculated for the 5-week period (dark grey) and the remaining time of the 10-week period (light grey).Horizontal arrows indicate the time of snowmelt in control plots (no snow manipulation), vertical arrows show the time of aboveground peak biomass harvest.Purple bars indicate artificial rain events (17 mm), where control plots (no rain-exclusion) were watered to counteract naturally occurring drought (two times in 2018, once in 2019 and 2022).the season (e.g.Klug-Pümpel, 1989;Körner, 2021;Wang et al., 2020).
Also, a transition from biomass to necromass (still attached to the plants) occurs after peak biomass and harvesting later would reduce the estimated annual biomass formation (Körner, 2021).From 2017 to 2022, we harvested above-ground phytomass in two 10 × 20 cm strips per plot that were not harvested before (Figure S1).Harvest took place during peak biomass usually within 1-2 weeks in early-to mid-August, except for 2022, when an extremely early season-start induced earlier growth and prompted us to harvest at the end of July.This practice implied that the harvested biomass experienced only 6-8 weeks of drought in the 10-week treatment and 5 weeks of drought plus 9-18 days of recovery in the 5-week treatment.However, the plants that were harvested were exposed to the full drought duration in all previous years, with potential legacy effects on biomass production.Freshly cut phytomass was stored at 4-6°C before sorting biomass to species-level.Each species was assigned to one of the functional groups of forbs, graminoids (comprising different grass-like species) and Carex.Although Carex curvula is a graminoid species, we treated it separately because of its dominance.Necromass was assigned to forbs and graminoids (including Carex).Leaves of Carex exhibit die-back from the tip downwards during senescence, leading to characteristic brown leaf tips that can be used as a proxy for senescence (Möhl et al., 2022).We cut these brown leaf tips of Carex at the yellow intersection to the green leaf and collected them separately as necromass in 2019-2022.Previous-year leaves, mainly of Geum montanum were assigned to necromass.In 2021, we quantified leaf area of 3-5 leaves per plot for five frequent species (Carex curvula, Helictotrichon versicolor, Geum montanum, Leontodon helveticus and Potentilla aurea).Leaves were fixed under a transparent acrylic plate and photographed beside a reference scale.Leaf area was extracted using Fiji (Schindelin et al., 2012) and specific leaf area (SLA) was determined after weighing oven-dried, single leaves.
Lichens and mosses were not accounted for in the above-ground biomass harvests.However, in 2022, total lichen mass was assessed by collecting all lichens within the 20 cm-respiration rings (see paragraph on ecosystem respiration), without differentiating between species, biomass, necromass and age (unfeasible for lichen).Once sorting was complete, all biomass and necromass samples were oven-dried at 80°C for at least 12 h and then weighed.

| Root biomass in in-growth cores
At the start of the growing season in 2017 and 2020, three ingrowth cores were installed per plot (Figure S1).We extracted soil cores to a depth of 20 cm (excluding the upper 1-2 cm of litter and raw humus with meristems) using a soil corer of 4.8 cm diameter.
Soil collected from the study site was sieved to remove all roots and filled into the holes (same soil for all in-growth cores).The refilled hole was covered with a vegetation patch of the same diameter.We sampled roots in 2018 and 2021, after two growing seasons each, by removing the vegetation patch and extracting the filled-in soil with a 4.4 cm (2018) and 3.8 cm (2021) corer at the end of the growing season in September, when seasonal root growth was complete (Möhl et al., 2022).We split the core in two to get the 0-10 and 10-20 cm horizons.Different volumes of the soil cores were taken into account when root biomass was transformed to g m −2 .

| Standing root biomass
To assess whether our treatment affected standing root biomass, we harvested two soil cores (4.4 cm diameter, 20 cm depth) of intact root space in the central-m 2 of each plot in August 2021 (Figure S1).
Cores were split into 0-10 and 10-20 cm and then stored at −20°C until processing.After thawing at 4°C, most roots were picked out with tweezers.Roots were then placed in a supersonic bath to loosen soil particles between fine roots and washed.The remaining soil-root mixture was sieved under running water (32 μm mesh size) to collect all other roots.Washed roots were oven-dried in the same way as above-ground biomass.Stone volume was determined by water displacement and subtracted from the soil volume before normalizing biomass to g m −2 .

| Community composition
We visually estimated plant cover (0%-100%) at species level in the central-m 2 of each plot in early August 2022.From the cover of vascular plants, we calculated two diversity indices: Shannon's index (Shannon, 1948) as an indicator of plant diversity (equally sensitive to rare and abundant species) and Simpson's index (Simpson, 1949) to assess changes in dominance (more sensitive to abundant species; Morris et al., 2014).Lichens and mosses were not determined to species level, but the majority of lichens belonged to Cetraria sp. and less frequently to Cladonia sp., while Polytrichum sexangulare Brid.
was the most abundant moss.

| Ecosystem respiration
Total respirational CO 2 (auto-and heterotrophic, 'ecosystem respiration') was measured with two identical static chambers (Bader & Körner, 2010;Mildner et al., 2015).They consist of a grey polypropylene housing with a 20 cm diameter opening.A battery-driven 3V-ventilator ensured gentle mixing of the air while CO 2 concentration and temperature were measured in the chamber (GMP343 and HMP75; Vaisala).Prior to the first measurement, each plot was permanently equipped with a polypropylene ring (height × diameter: 5 × 20 cm) in the central-m 2 (Figure S1).The ring was inserted ca.2.5 cm into the soil and served as a socket for air-tight chamber placement.Ecosystem respiration was measured in each plot during three measuring campaigns per season from 2019 to 2021.As measuring ecosystem respiration during the day can introduce significant error, we measured respiration during the night, starting 1 h after sunset and finishing 3-5 h later.After chamber placement on the socket, the increase in CO 2 was monitored for 5 min.The first minute of CO 2 readings was discarded to avoid anomalies caused by chamber placement.For the last measurement of 2021 we only used the last 2 min of CO 2 readings of one chamber, because the fan was not working properly and the increase in CO 2 stabilized only towards the end.Comparing respiration rates from the two devices did not indicate any measurement error and analysing the data with and without these measurements yielded the same statistical and qualitative outcome.Soil temperature at 0-5 cm depth was measured simultaneously at two positions beside the socket ring (GTH175; Greisinger).In 2021, we additionally measured ecosystem respiration in 10 spots within the experimental site but outside of the treatment plots.These spots were measured a second time after above-ground biomass was cut to estimate the proportion of above-and belowground respiration.Respiration rates were calculated from the linear regression slope of increasing CO 2 over time (ΔCO 2 , ppm s −1 ): where 22.4 (L mol −1 ) is the molar volume of any gas at standard pressure and temperature, 760 (hPa) is the local air pressure at the study site, 1013.25 (hPa) is the air pressure at sea level, 273.15 (K) is the standard temperature and T air is the air temperature in the chamber.V c is the chamber volume (5.65 L) and A the area (0.0314 m 2 ).Four single measurements were discarded because the regression had an R 2 of less than 0.9.As mean soil temperature per campaign varied between 7.8°C and 13.4°C (lowest value of 5.8°C and highest of 15.9°C), R was then standardized to 10°C (R 10 ) using the modified Arrhenius function by Lloyd and Taylor (1994): where R s is the respiration rate of the sample, T soil (in K) the concurrently measured soil temperature, and 283.15K equals to 10°C.E 0 (308.56K, 'activation energy') and T 0 (227.13K) are fitted parameters (Lloyd & Taylor, 1994).

| Statistical analyses
All statistical analyses were performed using the statistical programming language R 4.0.5 (R Core Team, 2022).We fitted mixed effect models (R-package 'nlme'; Pinheiro et al., 2021) with block as random effect.In the case of repeated measures, we also included plot identity as random effect.Where levels of a predictor showed substantially different variance (e.g.above-ground biomass of different functional types), we included constant variance functions that differed between factor levels ('weights' argument of the lme function).R 2 m -values reported for mixed effect models reflect marginal R 2 -values for fixed effects, derived with the package 'piecewiseSEM' (Lefcheck, 2016).Differences between treatments and species were computed using the R-package 'emmeans' (Lenth, 2021), which allows to perform t-tests on fitted values and thereby account for dependencies specified with random effects.Model assumptions regarding residual distribution were verified visually (quantile-quantile plots of residuals, residuals vs. fitted values).Multiple data points per plot were pooled prior to analysis (e.g. two harvest strips per plot, three ingrowth cores per plot and year).Correlations between below-ground biomass and total as well as species-specific aboveground biomass were investigated with least-absolute shrinkage (LASSO) regression ('glmnet'-package; Friedman et al., 2010), where coefficients of terms with low explanatory power shrink to zero.Resulting non-zero terms were then used to fit a linear mixed effect model.To assess whether above-ground biomass of the most common species (Table S1) responded to snowmelt timing and drought, we fitted individual regressions with 'year' and 'mean topsoil moisture between shelter placement and harvest' or 'day of snowmelt' as predictor variables, including interactions.Response variables were log or square-root transformed to improve the distribution of residuals in statistical models but back-transformed to report estimates on the original scale.If not stated otherwise, means ± SE are reported.

| Environmental conditions
Over the study duration, interannual variation in snowmelt date (2016-2022) spanned 42 days in control plots (2 July in 2020 and 21 May in 2022; Figures 1 and 2a).Snow removal led to an average advance of snowmelt by 4.6 ± 0.8 days and snow addition delayed snowmelt by 8.0 ± 1.9 days, with a maximum difference of 20 days between the two in 2020.Snow manipulations had minor effects on soil moisture, and only plots with delayed snowmelt were slightly moister (+2.1 ± 0.8 vol-%; Figure S2) during the first 5 weeks (p = 0.018) but not afterwards.Rain-out shelters excluded an average of 131 mm (32%-51% of the growing season's total) in the 5week and 232 mm in the 10-week drought treatment (69%-90%).
The drought treatment led to a steady decline in water availability (Figures S2 and S3).Soil moisture in the topsoil (0-5 cm) was reduced from 27.3 ± 0.7 vol-% in controls to 19.6 ± 0.7 vol-% in droughted plots (mean of the first 5 weeks), with a mean minimum value of 13.8 ± 0.6 vol-% (Figure 2b; no difference between the 5-and 10week treatment).During the second 5 weeks mean soil moisture dropped to 15.6 ± 0.6 vol-% in the 10-week treatment and recovered to 21.5 ± 0.7 vol-% in the 5-week treatment, where the rain-out shelters were already removed (Figure 2c).Still droughted plots had a mean minimum soil moisture of 13.2 ± 0.5 vol-% during that period.Mean soil moisture within treatments always remained above the permanent wilting point, although topsoil moisture dropped regularly below that threshold in individual plots under drought (Figure 2).Expectedly, differences in soil moisture were highest in the topsoil and declined with increasing soil depth (no difference at 40 cm depth, Figure 2d,e).From 2019 to 2022, we measured earlyseason soil moisture prior to the setup of rain-out shelters, which showed that the previous years' drought treatment left a signal in , the topsoil even right after snowmelt (−3.6 ± 0.9%, p < 0.001, no difference between 5-and 10-week).Mean soil temperature (at 3-4 cm depth) during the drought treatment varied between 11.0°C (2021) and 14.6°C (2022) in controls and was on average 0.7 K higher in droughted plots during the first 5 weeks (Figure S4).During the second 5 weeks, when soils were even drier in plots with rain-out shelters, this temperature difference increased to 1.2 K, suggesting less transpirational cooling under drought.Mean irradiance was 13.0 ± 3.1% lower under rain-out shelters.

| Above-ground biomass
Annual peak above-ground biomass in controls was 122 ± 6 g m −2 a −1 across all years and, aside from the first 2 years, biomass was remarkably similar between years (Figure 3).Total community biomass was unaffected by snowmelt date (Table 1) and including the interannual variation of snowmelt in the analysis by replacing snow treatment with the date of snowmelt in our model did not indicate a significant effect of snowmelt timing on total above-ground biomass production either (p = 0.78).Drought, however, decreased total biomass significantly in the 10-week treatment (−16.0 ± 7%, p = 0.011).
This effect was absent in the first year but then persisted during the following years (Figure 3).The response to the 5-week treatment varied significantly across years (Table 1), with a trend towards lower biomass in 2019 (p = 0.12) and a significant reduction in the final year 2022 (p = 0.008, Figure 3).This reduction in biomass was unrelated to the precipitation that the plants received between removal of the 5-week rain-out shelters and harvest, which amounted to 50-110 mm (and corresponding increases in soil moisture; Figure S3) in 2017-2020 but only <15 mm in 2021 and 2022.
Although total community biomass was not affected, snowmelt timing did exert differential biomass effects among functional groups (Table 1).Later snowmelt reduced biomass in graminoids (without Carex) by 9.6 ± 4.3% (p = 0.029), but had no effect on Carex or forbs.
When including the interannual variation of snowmelt in the analysis, the grass Helictotrichon was the only species of the 10 most abundant ones that was affected by snowmelt timing, with more than doubled peak biomass (from 4.2 ± 1.2 to 9.6 ± 2.1 g m −2 a −1 ) when snowmelt occurred at DOY 160 compared to 180 (p = 0.027, Figure S5).
Under drought, forbs and graminoids had lower biomass in the 10-week treatment, explaining the observed difference at the community level.In contrast, the dominant sedge Carex did not suffer from drought, and even produced slightly more biomass in the 5-week treatment (p = 0.031, Figure 3), increasing the proportion of Carex from 42 ± 2% in controls to 51 ± 2% in droughted plots (p < 0.001, no difference between 5-and 10-week).Regression analysis of the above-ground biomass in response to mean soil moisture between rain-out shelter placement and harvest confirmed these patterns, with less biomass in drier soil for the total community, forbs and  graminoids, but not Carex (Figure S6).Analysing the 10 most abundant species (together comprising >95% of total biomass) revealed that lower soil moisture significantly reduced biomass production in Geum, Helictotrichon and Agrostis, and marginally significant in Ligusticum (Figure 4).From the fitted models, we predicted average square-rooted biomass at 30 and 15 vol-% soil moisture and backtransformed the resulting values, which indicated a biomass reduction by 62 ± 21% for Geum, 54 ± 22% for Helictotrichon, 78 ± 24% for Agrostis and 50 ± 34% for Ligusticum.In contrast, biomass of Leontodon seemed to benefit from drier soil (+53 ± 32%; marginally significant, Figure 4).There was substantial variation in the lichen mass per plot, varying between 5 and 517 g m −2 (d.w.) with an average of 282 ± 36 g m −2 in controls.Drought significantly reduced lichen mass to 180 ± 31 g m −2 in the 5-week and to 130 ± 22 g m −2 in the 10-week treatment.
Necromass averaged at 38 ± 2.7 g m 2 a −1 in controls, varying between 26 and 52 g m 2 a −1 between years.There was marginally more necromass when snowmelt occurred earlier (42 ± 2.4 g m 2 a −1 , p = 0.061) and less under late snowmelt (31 ± 1.2 g m 2 , p < 0.001) across all years (Table S2).Drought reduced total necromass by 13 ± 5% (10-week only, p = 0.020), similar to the effect on aboveground biomass.Consequently, the ratio of necro-to biomass was unaffected by drought (p = 0.479).The response of necromass to snowmelt timing and drought was not significantly different between forbs and graminoids (including Carex here; Table S2).
Snowmelt timing had a similar effect on the dead fraction of Carex leaves as on total necromass: brown tips amounted to 22% of the green leaf mass in snow-controls, 26% under early (p = 0.003) and 16% under late snowmelt (p < 0.001).Drought also reduced this fraction by 3% (p = 0.003, no difference between 5-and 10-week).
Measured in five abundant species in 2021, SLA responded to drought only, with an average decline by 6.5 ± 2.6% in the 10-week treatment (Figure S7, p = 0.014, no interaction between drought and species).

| Below-ground biomass
The total amount of newly formed roots in in-growth cores (0-20 cm depth) was unaffected by snowmelt (Table 1) and amounted to 304 ± 14 g m −2 a −1 in controls, with ca.25% more roots at 0-10 cm soil depth compared to 10-20 cm (Figure 5).Drought increased root biomass in in-growth cores by 13 ± 6% in the 5-week treatment (p = 0.035) and this effect was independent of soil depth.Comparing the annual peak above-ground harvest with root growth in in-growth cores (both in g m −2 a −1 ) revealed an increase in the root mass fraction of (root mass / total mass) under drought (p = 0.009).
It was 70.7 ± 1.5% in controls and rose to 73.5 ± 1.4% in the 5-week and 75.2 ± 1.2% in the 10-week drought.Standing root biomass (incl.older roots and rhizomes) was influenced neither by drought nor by snowmelt timing (Table 1) and amounted to 1045 ± 36 g m −2 , with 762 ± 34 g m −2 in the upper and 283 ± 12 g m −2 in the lower soil horizon across all treatments (Figure 5).Fine roots with diameters <1 mm contributed ca.95% to total below-ground biomass, irrespective of treatment.Model selection using least-absolute shrinkage regression revealed that plots with more Carex above-ground biomass had significantly more below-ground biomass in the 0-10 cm soil layer (+9.5 ± 2.8 g below-ground biomass per g Carex above-ground biomass; p = 0.006), while total below-ground biomass was unrelated to total above-ground biomass.

| Community composition
Seven years of snow manipulation had no significant effects on plant cover, species richness, or diversity indices (Tables 1 and 2).
Also, 6 years of recurring drought had no effect in the 5-week treatment, but total plant cover was reduced by 12 ± 4% in the 10-week TA B L E 1 Linear mixed effect models testing for the effects of summer drought and altered snowmelt date on above-and below-ground biomass, plant cover and ecosystem respiration.drought treatment (p = 0.005; Table 2).This reduction was accountable to a decline in the cover of graminoids (−7 ± 2%, p < 0.001) and forbs (−9 ± 4%, p = 0.030).Plant cover of individual species reflected the species-specific declines in biomass production under drought (Table S1).In contrast, both drought levels had no negative effect on Carex cover, and the 5-week treatment even induced a trend towards higher cover (+6%, p = 0.10).Owing to the decrease in biomass of forbs and grasses, the relative contribution of Carex to total plant cover increased by 12 ± 4% in the 10-week drought treatment (36% in controls vs. 48% in 10-week; p = 0.003).Higher dominance of Carex was also reflected by the Simpson's index, which was significantly higher in the 10-week treatment compared to controls (p = 0.026; Table 2).In line with this, Shannon diversity decreased, although not significantly.Snowmelt and drought had no effect on moss cover, but lichen cover was nearly halved by the 10-week drought (from 6.9% to 3.6%; p = 0.001; Table 2).
Plots that melted out earlier than controls had a higher respiration in the two measurements conducted already in July, by 12 ± 6% in 2020 ) TA B L E 2 Estimated marginal means (± SE) of plant and lichen cover, species richness, and two diversity indices (Shannon's and Simpson's).' soil moisture (%-vol) later in the season (−3 ± 4% in August and +1 ± 6% in September, n.s.).
Thus, >85% of ecosystem respiration stemmed from below-ground activities, underpinning the overarching importance of respiration originating from below-ground biomass (including microbes).

| DISCUSS ION
We aimed at deciphering the effects of climate change on the productivity of this late-successional alpine grassland and employed a unique experimental approach where we combined manipulated snowmelt timing with recurrent summer drought over 6 years.Most importantly, recurrent drought reduced above-ground biomass, increased the root mass fraction and shifted the community composition towards the already dominant Carex curvula, which proved to be highly droughtresistant.In contrast, shifting snowmelt by 13 days on average between early and late snowmelt had little effects on any of the measured traits, underlining recent findings that snowmelt timing (and by that, season length) has little predictive value for biomass production in this type of alpine grassland (Möhl et al., 2022).Our results also highlight the differential controls over growth and reproduction: in contrast to growth, flowering phenology was mainly affected by snowmelt timing, while summer drought played a minor role in a previous study from the same experiment (Vorkauf, Kahmen, et al., 2021).

| Treatment effectiveness
Snow-removal and -addition in spring altered snowmelt timing by 13 days on average (maximum of 20 days), which represents a significant shift for the activity of alpine plants.However, this range is much smaller than the natural year-to-year variation in snowmelt (42 days), which is a characteristic of most snowmelt manipulation experiments (Rixen et al., 2021), as interannual variability in snowmelt is generally large in mountain areas (Vitasse et al., 2017).
Together, snow manipulation and inter-annual variation in the snowmelt date allowed us to test a large gradient of snowmelt timing of more than 50 days.Rain-out shelters effectively reduced soil moisture, with the strongest effects in the topsoil and observable differences down to 30 cm soil depth.This includes most of the rooted space in such grasslands (Greinwald et al., 2021;Hitz et al., 2001;Mezhunts et al., 2005).Beside precipitation, fog and dew formation during cold nights is a minor source of moisture, amounting to 2-6 mm per month (Groh et al., 2018;Riedl et al., 2021;van den Bergh et al., 2018), although rain-out shelters partly inhibit dew formation as radiative cooling during nights is reduced.

| Less biomass above-ground but more below-ground
Reduced above-ground biomass production in droughted grasslands has frequently been observed (Gao et al., 2019) across different land-use intensities and from low to subalpine elevation (Gilgen & Buchmann, 2009;Kröel-Dulay et al., 2022;Stampfli et al., 2018).As expected, our data show that alpine grassland is no exception, and peak above-ground biomass declined by 16% under recurring drought of 10 weeks per season.Interestingly, biomass was clearly not reduced by the 10-week drought in the first year of treatment, pointing towards drought effects on developmental processes after peak biomass (such as preformation of leaf buds) that might reduce biomass production in the following year.Because necromass was also lower under drought, the decline in biomass does not reflect increased or faster die-back but evidently a reduced biomass production.In contrast, a shorter drought of 5 weeks reduced peak biomass in the final year only, indicating some leeway for compensatory growth during drought recovery (Hahn et al., 2020).Yet, the time available for compensatory growth was obviously short, given that senescence and leaf die-back starts already 5-7 weeks after snowmelt in this vegetation type (Möhl et al., 2022).Topsoil desiccation during drought was more pronounced in 2022 and remained close to the permanent wilting point in the 5-week drought plots even during the recovery phase before harvest (Figure S3), possibly contributing to the lower biomass production in that year.The biomass decline we observed is less severe than that previously reported for a single-season drought experiment in a similar vegetation type, where 17 days of rain exclusion reduced above-ground production by 40% (de Boeck et al., 2015).However, this study used excavated monoliths that were decoupled from the soil continuum, which may explain the strong drought effect.Lower above-ground biomass under drought may counteract projected increases in growth due to warmer growing season temperatures (Rammig et al., 2010;Rumpf et al., 2022) and reduce food resources for various herbivores.
We also observed a massive reduction in the cover and mass of lichens (mainly Cetraria sp. and with lower abundance Cladonia sp.) under drought, although lichens are generally considered robust against harsh life conditions, including desiccation (Kranner et al., 2008).Very little is known about the effects of recurrent drought on ground-dwelling lichens in situ.We assume a massive dieback in Cetraria sp. but also increased dislocation by strong winds as dry thalli are less embedded in the less dense plant canopy under drought.
Low water availability can directly affect biomass production of plants by reducing turgor and impairing cell growth and biochemistry (Hsiao et al., 1976;Muller et al., 2011) but also indirectly by dampening nutrient uptake and availability in the soil (Da Silva et al., 2011).
Though soil moisture remained above the permanent wilting point for most of the experiment, we expect that both reduced water (through low turgor pressure) and nutrient availability caused much of the observed decline in biomass.It is known that growth of alpine grasslands is often limited by soil nutrients-in fact, a study conducted in the same vegetation type and close to our site found that above-ground biomass production increased by 50% when NPKfertilizer (45 kg ha −1 ) was applied (Schäppi & Körner, 1996).
The reduction of SLA in droughted plants along with reduced biomass and vegetation density (cover) in the 10-week drought treatment also points to a reduced leaf area index (LAI), contributing to community wide water savings (Chaves et al., 2003).A lowering of SLA in response to drought at high elevation had been observed earlier (Rosbakh et al., 2017), although a drought treatment with intermittent irrigation elicited only weak responses in SLA (Hamann et al., 2017).A meta-analysis of temperate grasslands at lower elevation found that SLA of grasses responds more strongly to drought than forbs (Wellstein et al., 2017).Yet, we could not confirm this pattern for the five alpine plants tested here.Another way to maintain the balance between water uptake and transpiration is to invest in below-ground tissues and improve access to soil water.
In general, drought stimulates root growth relative to shoot growth to balance resource acquisition (Bloom et al., 1985), especially in species-rich grasslands (Kahmen et al., 2005).Four seasons with 50% less precipitation resulted in nearly 30% more root biomass (in-growth cores) in alpine grassland on the Tibetan Plateau (Liu et al., 2021).Another study reported that below-ground biomass remained unchanged under drought, while above-ground production declined in subalpine grassland (Gilgen & Buchmann, 2009).
Whether or not below-ground biomass was enhanced, the root mass fraction increased in all these studies, similar to our findings.
A more productive alpine grassland, exposed to three seasons with 6-12 weeks of experimental drought showed an increased root biomass in in-growth cores by up to 40% in calcareous and 24% in siliceous sloping terrain (Schmid, 2017).Hence, we suppose that more frequent droughts in the future will further increase the already high root mass fraction in alpine ecosystems.Potential consequences include increased resource allocation to roots with unchanged or even reduced decomposition rates, which may increase root-derived carbon in alpine soils (Liu et al., 2021).
In our grassland, annual above-ground biomass productionapproximated by peak biomass-amounted to ca. 120 g m −2 , while below-ground production was 300 g m −2 to a soil depth of 20 cm, indicating that the large below-ground fraction of alpine grasslands is not only due to high root longevity but also a result of substantially more below-ground investments.Notably, we compared peak above-ground harvests with end-of-season below-ground harvests because roots grow for a much longer period than leaves in this ecosystem and harvesting in-growth cores together with peak above-ground biomass may underestimate annual below-ground production by up to 50% (Möhl et al., 2022).
We aimed at identifying longer-term effects of climate change on biomass formation, but did not observe any duration-related trends over 6 years of combined snow manipulation and drought treatments.Biomass responses were either fairly stable or fluctuated between years with no clear trend.We assume that the absence of additive effects of recurring drought is a consequence of the large influx of water during snowmelt, which largely reset the system.Nonetheless, we found that control plots were moister than droughted plots right after snowmelt in the last 3 years, suggesting a legacy effect, namely that drought reduced soil field capacity over the course of the experiment.This implies changes in soil structure and properties under recurring drought, for instance, increased water repellency (hydrophobicity) due to a change in the structure of soil particles (Goebel et al., 2011;Robinson et al., 2016), which could affect nutrient relations and plant growth in the long term under more frequent drought.

| Reduced ecosystem respiration under drought
We observed efflux rates of 1-2 μmol CO 2 m −2 s −1 during the growing season, which is low compared to lower elevation grasslands (e.g.Burri et al., 2018: 4-8 μmol CO 2 m −2 s −1 ), but typical for high-elevation vegetation with low soil temperature and productivity (Bahn et al., 2008;Diemer, 1994).Year-round eddy covariance measurements at a nearby alpine grassland site yielded ecosystem respiration rates in a similar range (0-5 μmol CO 2 m −2 s −1 during the growing season; Scholz et al., 2018).The biggest fraction (>85%) of respiration was accountable to below-ground activity, as cutting aboveground tissues reduced respiration by no more than 15%, which is nearly equal to the portion of total biomass (incl.roots) made up by above-ground tissues in our experiment (14.2% in controls).
As expected, drought resulted in consistently lower ecosystem respiration in plots with rain-out shelters compared to controls, implying that reduced above-ground biomass and dry soil constrain plant and microbial activity (Hasibeder et al., 2015;Schimel, 2018).
However, given that drought did not reduce root biomass in our case, we hypothesize a stronger negative drought effect on soil microbes than on plant organs.Such results are consistent with reduced soil respiration in other drought experiments in lowland (non-alpine) grassland ecosystems (e.g.Bloor & Bardgett, 2012;Burri et al., 2018;Harper et al., 2005) and in managed subalpine grassland (Hagedorn & Joos, 2014;Ingrisch et al., 2018).While droughted plots respired less during each of the nine measurement campaigns between 2019 and 2022, we might have missed ephemeral respiratory bursts due to rewetting immediately after removal of the rain-out shelters (Barnard et al., 2020;Ingrisch et al., 2020;Schimel, 2018).Such bursts can result from rapid mineralization of litter and soil organic matter upon rewetting and may, beside stimulating microbial growth, also increase plant nutrient uptake and growth (Karlowsky et al., 2018).
Especially in the 5-week drought treatment, such a burst of nutrients due to rewetting may have contributed to biomass recovery.CO 2 efflux rates during recovery from drought increased to the level of controls but never exceeded them.A 3-year drought study in managed subalpine grassland measured soil respiration at higher temporal resolution but found no indications for higher respiration after rewetting (Hagedorn & Joos, 2014).On the other hand, respiration was stimulated by rewetting after a 6-week drought in managed and abandoned subalpine grassland (Ingrisch et al., 2020).It awaits further research to explore whether the reduced rates of respiration under drought are compensated during wetter periods, particularly during the 8-9 months lasting winter period at our site.

| Weak effects of snowmelt timing
We hypothesized that earlier snowmelt could lead to higher aboveground biomass after several years.To our surprise, even over 7 years of snow manipulation, altered snowmelt timing had hardly any effects on the vegetation.Several studies in arctic tundra and alpine vegetation observed a similar unresponsiveness in productivity to earlier snowmelt, both above-and below-ground (Baptist et al., 2010;Boyle et al., 2022;Darrouzet-Nardi et al., 2019).It was previously argued that adverse effects of freezing temperatures during earlier snowmelt outweigh beneficial effects of a longer growing season (Wipf et al., 2009) but particularly graminoids exhibit a high freezing resistance even in a fully active state (von Büren & Hiltbrunner, 2022).In addition, internal controls over growth and senescence could restrain plants from extending growth when the season prolongs (Körner et al., 2023), as recently evidenced for alpine Carex curvula grassland (Möhl et al., 2022).
Nevertheless, a remote sensing study on alpine grasslands between 2000 and 2600 m elevation in the Alps found a correlation between growing season length (snowmelt timing) and NDVI-derived gross primary production (Choler, 2015).However, the higher integrated NDVI signal in seasons with earlier snowmelt stemmed mainly from an extension of the senescence period, as peak NDVI always occurred ca.
50 days after snowmelt.This indicates that remaining green tissues, rather than new growth, were responsible for the observed correlation.In absence of direct effects, the timing of snowmelt is assumed to influence plant growth mainly through changes in community composition, as plant species (other than Carex) with already longer growing phases may become more competitive in a future climate with much earlier snowmelt.While the average shift in snowmelt of 13 days was not enough to affect community composition, the substantial advance in snowmelt of up to 1 month projected for the end of the 21st century (Vorkauf, Marty, et al., 2021) will push snowmelt largely out of the range of its current interannual variation, and could therefore exert greater pressure on the prevailing plant community.Although the Alps have become greener (higher NDVI) at montane and alpine elevations during the last four decades, this greening was associated with warmer temperatures rather than a decline in snowcover during summer months (Rumpf et al., 2022).
Manipulated snowmelt timing also had no effect on the drought treatment, except for the slightly higher soil moisture in the first 5 weeks under delayed snowmelt.It appears that the consequences of the two treatments were largely independent and that our drought-related findings are representative of future conditions with earlier snowmelt.One reason for the consistency of drought effects among snowmelt treatments could be that we initiated both drought levels immediately after snowmelt.Growth of alpine plants tightly tracks snowmelt (Choler, 2015;Vitasse et al., 2017), and starts whenever temperature is suitable for growth.Consequently, the main phase of biomass production always occurs in the first weeks of the season (Möhl et al., 2022) and drought is expected to affect biomass production most during this phase.This also means that drought later in the season may affect productivity of alpine grassland to a lesser extent or only for species that still grow later in the season (e.g.Helictotrichon versicolor or Potentilla aurea; Möhl et al., 2022).

| Species identity matters
Graminoids and forbs differ in many aspects such as plant architecture, metabolism, or dispersal strategy (Wullschleger et al., 2014) and therefore respond differently to environmental change.For example, a study on the Tibetan Plateau reported stronger effects of drought on the abundance of forbs and sedges, while the abundance of grasses (Stipa aliena, Elymus nutans, Helictotrichon tibeticum and others) increased over 4 years of experimentally reduced precipitation (Liu et al., 2018).In our study, biomass of both forbs and graminoids other than Carex declined in the 10-week drought treatment.
It is striking that biomass production of the dominant species, Carex curvula, was not affected by drought and even slightly increased in the 5-week drought treatment.After 6 years of treatment the dominance of Carex in droughted plots was indeed higher.Therefore, the consequences of drought in this alpine grassland appear to be specific to species rather than to functional groups and thus, depend substantially on the local species pool.
Carex curvula appears to have a very conservative growth control, with above-ground biomass hardly responding to varying environmental conditions such as elevated CO 2 (Schäppi & Körner, 1996), shading (Möhl et al., 2020) and as revealed here to summer drought.
In part, low growth rates, preformed leaf buds and tiller endurance of several years may confer this high resistance to variable life conditions.Its below-ground dominance through substantial root longevity (Budge et al., 2011;Leifeld et al., 2015) may considerably add to the high drought tolerance (Mähr & Grabherr, 1983), which fits our observation that standing root biomass correlates with Carex cover.Interestingly, Carex curvula roots are known to be heavily colonized by dark septate fungal endophytes (DSE), quite a different group of fungi than those forming VA mycorrhizas (Read & Haselwandter, 1981).Whether these dark-septate fungi mediate the enhanced drought resistance in this species has not been explored so far.DSE isolated from roots of an epiphytic orchid and inoculated to seedlings of the same species increased their drought resistance (Liu et al., 2022), underpinning the role of DSE for the hosts.
In contrast to Carex, two of the most abundant species-the forb Geum montanum and the grass Helictotrichon versicolor-which together accounted for about 30% of total biomass production, turned out to be sensitive to drought.A study with 11 alpine species from the Rocky Mountains and the Sierra Nevada grown from seed revealed differential growth responses to drought, with those species from drier sites being less affected (Peterson & Billings, 1983).Although Geum and Helictotrichon are not known for preferring moist habitats (both have an Ellenberg moisture indicator of 5; Ellenberg & Leuschner, 2010), water shortage may render them less competitive within this plant community, reducing their abundance under future drought scenarios.In contrast, the low sensitivity to drought suggests that Carex will become even more dominant in the future.

| CON CLUS IONS
Despite our expectations, we found almost no evidence of interactive effects between snowmelt timing and summer drought in this ecosystem.Our results show that several weeks of summer drought reduce biomass production of this alpine grassland, with consequences for grazers on managed-and unmanaged land under continued climate change.While most of the biomass of alpine grassland is already below-ground, the proportion of root mass will further increase in the future should drought conditions become more frequent, underscoring the need to understand root and soil processes as well as their interaction in these ecosystems.Drought appears to have hampered soil microbial activity in the most active soil horizon, which could limit nutrient and carbon cycling and lead to temporal discrepancies between nutrient availability and plant demand.We find it remarkable that experimental drought increased the dominance of Carex curvula in only 6 years, despite the recognized persistence and longevity of such dense, late-successional vegetation with predominantly clonal growth (Vittoz et al., 2009;Windmaißer & Reisch, 2013).A major outcome of this study is therefore that the dominant species profits from the simulated scenarios, perhaps explaining the several millennia of persistence of individual Carex clones in this grassland (de Witte et al., 2012).Given the unresponsiveness of biomass production to variable snowmelt dates, our study warns against projections that assume proportionality between biomass production and season length (see Körner et al., 2023).In contrast to drought, systematic variation of snowmelt date did not affect species composition, a surprising result in view of the classical species gradients across snowmelt gradients.Either, late successional communities represent an end point of species assemblage that is far more robust than the species composition of snowbeds, or such 'timing effects' take longer than drought effects to materialize, given the natural variation of snowmelt timing.
Each plot was assigned to one of three levels of snow manipulation and summer drought for the whole duration of the experiment (3 × 3 full factorial block design).Snow manipulations were conducted each spring from 2016 to 2022, a few weeks before snowmelt to prevent shovelled holes from being refilled by recent snowfall.Remaining depth of the melting snowpack amounted to ca. 1 m at this time.Snow height was experimentally reduced to 0.5 m to advance snowmelt and elevated to 2.2-2.5 m (2016; 2017) or 1.3 m cm depth in the centre of the plots (Profile probe PR2; Delta-T Devices; for simplicity referred to as 10, 20, 30, and 40 cm depth).To quantify reductions in irradiance due to the rain-out shelters, we placed light loggers (HOBO Pendant Temperature/Light Logger; Onset Computer Corp.) in three controls and three sheltered plots from 27 August until 23 September 2021, when shelters were removed.All light loggers remained in the field until 29 September to measure a reference period for standardization of logger readings.
temperature (°C) Precipitation (mm)/Soil moisture (vol-%) Day of year of snowmelt (2016Day of year of snowmelt ( -2022) )  and (b, c) mean topsoil (0-5 cm) moisture for each experimental year (2017-2022, first rain-out shelters in 2017).Small points depict values for each plot and bigger points show means ± SE for each treatment.DOY of 140 is equal to 21 May, DOY of 200 to 20 July.Topsoil moisture is split in two periods: from 0 to 5 weeks plots of both drought treatments (5-and 10-week) were sheltered from rain, while from 5 to 10 weeks only the 10-week plots were sheltered.Dashed line indicates soil moisture at the permanent wilting point.(d, e) Difference in soil moisture between the two drought treatments and controls (horizontal black line) across soil depths, averaged over all seasons (FigureS3for continuous soil moisture values at 0-5 and 10 cm depth).
(p = 0.054) and by 21 ± 6% in 2021 (p = 0.002).This effect disappeared later in the season.Over all dates, drought on average reduced ecosystem respiration by 13 ± 5% in the 5-week (p = 0.015) and by 23 ± 5% in the 10-week treatment (p < 0.001).The strength of this effect varied between measurement dates, depending on whether plots were still under drought.Under rain-out shelters, the deviation from controls increased with drought duration (up to 40%; p < 0.001, R 2 = 0.80; Fig-ure 6d) and as the difference in soil moisture content became more pronounced (p = 0.03, R 2 = 0.49; Figure6e).Removal of above-ground foliage before the measurement reduced ecosystem respiration by 13 ± 4% in early season of 2021 (July), but this difference disappeared F I G U R E 4 Species-specific soil moisture sensitivity of annual peak biomass in the 10 most frequently occurring species (ordered alphabetically within functional groups).Percentage numbers in brackets indicate the fraction of biomass per species across all control plots.Points indicate regression slopes of peak biomass (square-rooted, √ g m −2 a −1 ) against mean topsoil moisture (0-5 cm, vol-%) between the placement of rain-out shelters and harvest.Negative values (to the left) indicate more biomass in drier soil and vice versa.Small points depict slopes for each of the 6 years, while bigger points show the mean ± SE across all years.Marginal R 2 -values are reported.Below-ground biomass in response to drought treatments.Left: Fine root biomass of in-growth cores, collected after 2 years of incubation (2017/2018 and 2020/2021).Right: Standing below-ground biomass (note different y-axis dimension) including fine roots (>95% of total), thicker roots (>1 mm) and rhizomes, collected in 2021.Total bar size indicates mean ± SE per treatment.The upper and lower part of each bar shows the biomass at 0-10 cm and 10-20 cm soil depth, respectively.

F
I G U R E 6 (a-c) Ecosystem respiration in response to 5-and 10-week of recurring summer drought, normalized to 10°C soil temperature.Coloured background indicates the sheltering periods of the 5-and 10-week drought treatment.Light-coloured points indicate values for each plot and dark-coloured, bigger points show means ± SE for each drought treatment (no interaction with snowmelt, n = 15).Points of the three treatments are slightly shifted to improve visibility.(d, e) Respiration (mean, 95%-CI in grey) in droughted plots as a fraction of control plots across all three seasons.The fraction declines with increasing drought duration (d) and/or with larger soil moisture difference between control and droughted plots (e).
In the case of plant cover, estimates and standard errors were back-transformed from the square-root values, meaning that cover by functional group (FG) does not sum up exactly to total plant cover.Values in bold differ significantly (p < 0.05) from controls.