Research site selection and basic site information
The study was conducted in the Hengduan Mountains, Sichuan Province, China (Fig. 1). This area is on the south-eastern edge of the Tibetan Plateau, and the study sites are located in Kang-Ding Valley, on the northwest slopes of Mt. Gongga along a steep elevational gradient where the lower part is characterized by mixed coniferous-broadleaved forest, followed by belts of subalpine coniferous forest, sub-alpine meadows, subalpine shrub, and alpine meadows towards higher elevations and colder climates33,34.
In 2012, four study sites along the slope were chosen to reflect the major bioclimatic variation in southwest mountain grassland vegetation along an elevation gradient of 1100 meters; the Lowland site at 3000 m a.s.l, the Middle sites at 3500 m a.s.l., the Alpine site at 3850 m a.s.l., and the High Alpine site at 4130 m a.s.l. (Fig. 1, Table 1). All the sites were placed selectively in grasslands associated with mountain gray-brown soil originating from granite35. The sites were selected to represent the charactieristic grasslands at different elevations and hence climatic conditions, but to otherwise be as similar as possible with respect to environmental conditions, vegetation structure, plant community composition, slope, aspect, etc. The geographical distance between adjacent sites is on average ca 2 km.
At each site, we selected an experimental area within the grassland as homogeneous and representative of the grasslands at that elevation as possible. The experimental areas were placed on sloping ground, avoiding depressions and concave areas in the landscape and other features such as big rocks or formations that may affect microclimate, light conditions, hydrology and/or snowdrift. All sites were moderately grazed prior to the experiment by yak, sheep, cattle, goats, and/or horses, and the experimental areas were fenced and locked for the duration of the study to prevent grazing and human disturbance of the experimental infrastructure. The fenced area was mowed at the end of each growing season to mimic past grazing regimes and minimize fence effects.
Block and plot setup
Within each of the four fenced experimental areas, we established seven replicate blocks, with a distance between the blocks ranging from four to six meters. Blocks were selectively placed in homogenous grassland, avoiding rocks, depressions, and other features as described above. Each block comprised five 25 cm × 25 cm plots, which were placed in a regular 3 rows x 2 columns grid (facing uphill) with 50 cm between adjacent plots. If a plot contained more than 10% bare rock, shrubs, or other non-grassland features, they were rejected or moved slightly to avoid these features. The plots were numbered by row from the uphill left corner of each block, permanently marked with plastic poles in each corner, onto which a standardized vegetation analysis frame could be mounted for vegetation (re)sampling. Within sites, the experimental area containing all experimental blocks is situated within a total area of ca. 75–200 m2.
In 2015, 73 additional 50 cm × 50 cm plots were set up within the fenced block areas in between the experimental blocks for biomass sampling (see dataset iii).
Transplant and Open Top Chamber) experiments
Within each of the seven blocks at each site, the pre-established and numbered plots were randomly designated to the following experimental treatments (with the specific treatments depending on the site, Fig. 1): (1) passive warming with an Open Top Chamber (OTC), (2) transplanting to a site one step warmer along the gradient (treatment ‘Warming’; from sites High alpine, Alpine, Middle), (3) transplanting to a site one step colder along the gradient (‘Cooling’; from sites Alpine, Middle, Lowland), (4) transplanting down the entire gradient (‘Extreme warming’; from site High alpine), and (5) up the entire gradient (‘Extreme cooling’, from site Lowland), (6) transplanting within blocks (to control for the transplanting itself)(‘Local transplant’; all sites), and (7) an untouched control plot (‘Control’; all sites). Thus, each OTC has a local unmanipulated control, and each transplanted turf has an “origin” site and a “destination” site, each with two types of controls, a local transplant and an untouched control plot (Fig. 1). The OTC chambers are hexagonal, 40 cm tall, with a distance between parallel sides of 106 cm at the base and 60 cm at the top. Snow cover is low and scattered at all sites, and the OTCs were left out year-round to achieve warming during the entire snow-free period. Because of the low snow cover, removing OTCs was also not necessary for their protection against snow damage.
We marked the upslope centre of each plot with a plastic flag to ensure that all plots were always analysed from the same direction, and that transplanted turfs retained their orientation relative to the slope and block layout at the destination site. For the transplant treatments, we used a knife to cut the turfs 2 cm outside the plot margins, giving turfs of 29 cm × 29 cm and to a depth of 20 cm, unless the organic soil was shallower, as was the case for some of the higher-elevation plots. After excavation, the turfs were packed into 29 cm × 29 cm waterproof boxes and transported to their respective destination sites within one or two days. To keep the transplant disturbance as similar as possible among treatments, the excavated control (‘home transplant’) turfs were also kept in boxes before they were put into their designated randomized destination plots within the site of origin.
The transplanted turfs were fitted into gaps created by excavating turfs at the destination site. Each block received one plot of each relevant treatment, resulting in five turfs per block, but with different specific treatments in the two middle sites and the extremes of the gradient (see Fig. 1). Transplanted turf positions were randomized within the destination blocks, while un-transplanted control and OTC warming plots – were not moved. Transplanted turfs were then carefully planted into their destination plots ensuring that the original orientation of the turf vs. the block slope was retained, the soil surface was level with the surrounding soil surface, and the edges of the excavated plot were in good contact with the sides of the gap. If necessary, loose soil was carefully removed from the underside of the turf, or local soil was added to the gap or around the edges to achieve this. This experimental design had a total of 140 plots. In 2014, two replicate blocks (10 plots) in the High alpine site were damaged by yaks, so having only five blocks in this site going forward.
Species identification, taxonomy, and flora
All species sampled in the vegetation and functional trait datasets were identified in the field. Back at the field station, the identification of sampled whole plants or vouchers were checked by a taxonomic expert from the region using the Online Flora of China36. Specimens problematic to identify were brought back to the Chongqing Normal University for identification and deposition of vouchers by one of the co-authors (Prof. He). Forbs were identified to species level, whereas many of the graminoids were identified only to genus level, i.e., Carex spp., Poa spp., Kobresia spp., and Festuca spp., due to difficulties with identification of sterile graminoids as there may be undescribed taxa and as there are no keys for vegetative plants for this region (professor He, personal observation). All taxon names were standardized using the Taxonomic Name Resolution Service37.
Dataset collection methods
Dataset (i): Plant community composition sampling
All vascular plant species in each plot were surveyed in 2012 (before treatment), and annually between 2013 and 2016. Each year, vegetation was surveyed during the peak of the growing season using a 25 cm × 25 cm frame overlain with a grid of 5 cm × 5 cm subplots. Subplots were numbered 1–25 starting from the up-slope left-hand corner of the plot, numbering the subplots from left to right by rows. We registered presence-absence of all species in each subplot, and estimated the percentage coverage of each species in the whole plot to the nearest 1%. Note that the total coverage in each plot can exceed 100, due to layering of the vegetation.
Dataset (ii): Vegetation height and structure sampling
Vegetation structure data for each plot was recorded between 2012 and 2016. Mean vegetation height and bryophyte depth was measured at five evenly spaced points per plot using a ruler. The total percent coverage of all vascular plants combined, was also recorded.
Dataset (iii): Biomass sampling
We measured the standing biomass per vascular plant species in the 73 additional 50 cm × 50 cm plots placed out in the general experimental area of each site in 2015. For each plot, species composition was registered, percentage cover of each species estimated using a 50 cm × 50 cm frame with 10 cm × 10 cm overlay, and the height measured at five positions using a ruler. The plot was then harvested at the ground level, sorted to species, and the biomass dried at 65 °C for 3 days before weighing to the nearest 0.0001 g. As the aboveground grass and forb biomass dies back each winter, standing biomass can be considered an approximation of aboveground net primary productivity in these grasslands.
Dataset (iv): Trait sampling and lab analyses
Site-level sampling for leaf trait analyses. We collected whole plants for leaf trait analyses from the common species in the plant community at each of the four sites in August 2015 and 2016. In 2015, we sampled as many species as possible from each of the four sites. In 2016, the collection was complemented to ensure that we had trait data from species making up at least 80% of the vegetation cover in all control plots at each site. The plants were collected outside of the experimental plots within a 50 m perimeter from the blocks, and we aimed to collect up to five individuals from each species in each site. To avoid repeated sampling from a single clone, we selected individuals that were visibly separated from other ramets of that species. The sampled plant individuals were labelled, put in plastic bags with moist paper towels, and stored in darkness at 4 °C until further processing. Processing was done as soon as possible, but due to unexpected high diversity in the grasslands and therefore many collected plants, some specimens were stored for longer than optimal, in 2015 for up to 4 days and in 2016 for up to 2 days. As a result, some of the site-level samples, especially from 2015, were less well hydrated/more wilted and/or generally in a poorer state than what is optimal for trait measurements (see below for discussion of data documentation and quality checking; note that the raw leaf scans are available for all leaves). Before processing, plant identification was checked vs. the Online Flora of China36. Up to three healthy, fully expanded leaves were then sampled from each individual. The leaves were cut off as close to the stem as possible, including the blade, petiole, and stipules when present. Further processing was completed within 24 hours (see below).
Experimental treatment-level leaf sampling for trait analyses
Ten of the most common species along the gradient were selected for more intensive sampling within the experimental treatments plots in August 2016. We targeted species (i) with broad distributions along the gradient, (ii) that were also locally frequent and therefore present in a number of turfs prior to the experiments, and (iii) that remained present in most treatments until the sampling year. This was to ensure sufficient replication within species across sites and treatments. Further, to prevent measuring individuals that had colonized the sampled plots during the course of the experiment, and therefore had not been subjected to the experimental treatments, we (iv) excluded species that spread fast clonally. The selected species for treatment-level analysis are Artemisia flaccida, Epilobium fangii, Geranium pylzowianum, Hypericum wrightianum, Pedicularis davidii, Polygonum viviparum (Persicaria vivipara), Plantago asiatica, Potentilla leuconota, Veronica szechuanica, and Viola biflora var. rockiana. Of these species, P. leuconota and V. szechuanica were present across the whole elevation gradient. All other species were sampled from all treatments at sites where they are naturally occurring, including from transplanted turfs outside of the natural elevation range. Because many species in this system display at least some clonal reproduction, distinguishing genetic individuals is impossible without destructive sampling, and we therefore work at the ramet level, following38. For all sites and selected species, we collected up to five healthy, fully expanded leaves from up to five ramets in all experimental plots (i.e., control, locally transplanted control, moderate and extreme warmed and cooled transplants, open top chamber) where that species occurred. The leaves were cut off as close to the stem as possible, including the blade, petiole, and stipules when present. The leaves were labelled, put in plastic bags with moist paper towels, and stored dark at 4 °C and further processed within 24 hours of collection.
Plant functional trait measurements
We measured 11 leaf functional traits that are related to potential physiological growth rates and environmental tolerance of plants, following the standardized protocols in Pérez-Harguindeguy et al.39: leaf area (LA, cm2), leaf thickness (LT, mm), leaf dry matter content (LDMC, g/g), specific leaf area (SLA, cm2/g), carbon (C, %), nitrogen (N, %), phosphorus (P, %), carbon nitrogen ratio (C:N), nitrogen phosphorus ratio (N:P), carbon13 isotope ratio (δ13C, ‰), and nitrogen15 isotope ratio (δ15N, ‰).
Initial leaf processing was done at the field station of the Institute for Mountain Hazards and Environment (IMHE) in Moxi, Sichuan. Processing was done in the following steps:
Leaf area. Leaves (including blade, petiole, and stipules when present) were carefully patted dry with paper towels, flattened (folded out to their maximum area) and scanned using a Canon LiDE 220 flatbed scanner at 300 dpi. Care was taken that no leaf parts were overlapping on the scanner, and naturally overlapping parts of lamina (e.g., as is the case in some compound leaves) were cut off and placed next to each other on the scanner to obtain the full leaf area. Leaves that could not be flattened because they are narrow and grow naturally folded (e.g., some Festuca sp.) were scanned as they were, thereafter the area was multiplied by 2 during data processing. Any dark edges on the scans were manually cropped. Leaf area was calculated using ImageJ40 and the LeafArea package41.
Leaf wet mass. Each leaf (including blade, petiole, and stipules when present) was weighted to the nearest 0.001 g to assess fresh mass.
Leaf thickness. Leaf thickness was measured at three locations on each leaf blade with a digital calliper (Micromar 40 EWR, Mahr) and the average was calculated for further analysis. When possible, the three measurements were taken on the middle vein of the leaf and on lamina with and without veins. The petiole or stipule thickness is not measured.
Leaf dry mass. Leaves (including blade, petiole, and stipules when present) were then dried for a minimum of 72 hours at 65 °C before dry mass was measured to the nearest 0.0001 g.
Leaf stoichiometry and isotopes. A subset of leaves (n = 576; 265 from the gradient sampling and 311 from the experiments) were transported to the University of Arizona for leaf stoichiometry and isotope assays (P, N, C, δ15N, and δ13C). The leaves were stored in a drying oven at 65 °C before shipping and processing. Each leaf (including blade, petiole, and stipules when present) was ground into a fine homogenous powder. Total phosphorus concentration was determined using persulfate oxidation followed by the acid molybdate technique (APHA 1992) and phosphorus concentration was then measured colorimetrically with a spectrophotometer (ThermoScientific Genesys20, USA). Nitrogen, carbon, stable nitrogen (δ15N) and carbon (δ13C) isotopes were measured by the Department of Geosciences Environmental Isotope Laboratory at the University of Arizona on a continuous-flow gas-ratio mass spectrometer (Finnigan Delta PlusXL) coupled to an elemental analyzer (Costech). Samples of 1.0 ± 0.2 mg were combusted in the elemental analyser. Standardization is based on acetanilide for elemental concentration, NBS-22 and USGS-24 for δ13C, and IAEA-N-1 and IAEA-N-2 for δ15N. Precision is at least ± 0.2 for δ15N (1 s), based on repeated internal standards. Ratios between C:N and N:P were calculated and used in the analysis.
Datasets (v-vii) – climate data
Climate stations (U30-NRC, Tempcon Instrumentation LTD, UK) at each site recorded precipitation, air temperature and humidity at height of 2 m above ground, and soil moisture were measured at depths of 0 cm, 5 cm and 20 cm below ground in 10 min intervals, starting in September 2012 and continuing during the whole study period. The same measurements were conducted inside OTCs except that here air temperature was measured at 20 cm above ground. Additionally, iButtons (iButtonLink Technologies, USA) were installed at 5 cm below ground, ground level, and 30 cm above ground inside and outside OTCs in each site along the gradient from June–August 2017, at 10 min intervals. Finally, TomsT TMS-4 loggers42 were installed inside and outside OTCs at the High alpine site from September to November 2019. These measure temperatures at 15 cm above ground, ground level, and 6 cm below ground, and soil moisture, at 15 min intervals.