Materials and Methods
This study was conducted in the town of Xihai, Haiyan County (100°57′N, 36°56′E, 3100 m ASL), Qinghai Province, China, at the eastern edge of Qinghai Lake (Fig. S1). Mean annual temperature is 1.4°C, mean annual precipitation is 330–370 mm, and mean annual potential evaporation is approximately 1,400 mm. The typical vegetation is alpine meadow, dominated by sedges and grasses. The soil type is loam, and the vertical profile is 0–30 cm for clay loam and 30–50 cm for silty loam.
Four key species in the alpine meadow on the QTP, i.e., Stipa purpurea (grass), Kobresia humilis (sedge), Artemisia scoparia (non-legume forb), and Astragalus laxmannii (legume) have been monitored from 1997 to 2017 (with one observation for each of the four species in 2000; for 2006, data for Astragalus laxmanniiwere missing) (Figs. S1 to S5). A randomly selected area of 100 × 100-m was equally divided into four 50 × 50-m blocks; each block was divided into four 25 × 25-m cells and marked permanently, and each cell was further separated into three sections. The first section was used for monitoring phenological time periods (e.g., green-up, flowering, fruiting, and withering), using 10 individuals of each species in each cell. Plant phenology was recorded every 2 days after the green-up period. The second section was used for monitoring plant height, using 10 individuals of each species in each cell. Plant height was recorded every 10 days after the green-up period. The last section was used for monitoring plant biomass, for which each plant in a randomly selected 1 × 1-m quadrant was harvested after the withering period, and the selected cell was rotated clockwise every 4 years in each block (Fig. S1). After harvesting, each plant was oven-dried at 65°C to constant weight. Soil moisture at a depth of 0–30 cm was determined via oven-drying of soil samples at 105℃ to constant weight every 10 days from March to the end of September. The climatic data were obtained from the Haiyan meteorological station (nearby the study site), recording daily average temperature, daily maximum temperature, daily minimum temperature, daily average soil temperature (0–30 cm), daily precipitation, daily relative humidity, and daily sunshine length (note: data of daily sunshine length were missing in the second half of 2017).
We defined the date when the phenomenon occurred (Julian day) in more than 50% of the marked plants as the time of the phenological period (e.g., green-up, flowering, fruiting, and withering) and the duration (number of days) from flowering to fruiting as the length of the generative growth period. We used the duration (number of days) from green-up to withering minus the length of the generative growth period as the length of the vegetative growth period.
To describe the growth dynamics across growing seasons with the height measured every 10 days (Figs. S2 to S5), we used a three-parameter logistic function (Paine et al. 2012). We also calculated the annual peak height, the timing of maximum growth (Julian day), and the intrinsic growth rate. To determine the start and end growth for each species, we consulted previous studies (Richardson et al. 2013) and set the time to reach 21% of the maximum height as the time for the start of the rapid growth phase (Julian day) and to reach 79% of the maximum height as the time for the end of the rapid growth phase (Julian day). We defined the duration (number of days) from the beginning of rapid growth to the end of rapid growth as the length of the rapid growth phase.
To determine phenology dynamics, growth patterns, and annual biomass in different plants, we used variance analysis (Table S1) and employed a linear model to show the trends at the start of the rapid growth phase, the timing of maximum growth, the end of the rapid growth phase, green-up time, flowering time, fruiting time, withering time, annual peak height, intrinsic rate of plant growth, and biomass of key plant species (Figs. 1 to 3). We used linear regression to explore the relationships among plant biomass, intrinsic growth rate, length of the rapid growth phase, length of the vegetative growth period, and length of the generative growth period over time (Table S2).
To identify optimal length thresholds (i.e., how many days) and the factors determining the growth and phenological events which are significantly associated with plant biomass, we also used linear models to evaluate the effects of average temperature, maximum temperature, minimum temperature, average soil temperature, precipitation, relative humidity, soil moisture, and sunshine length on plant growth (start of the rapid growth phase, the timing of maximum growth, and the end of the rapid growth phase) and phenological events (green-up, flowering, fruiting, and withering) during the growing season. To determine which days affected the key growth and phenological events, we compared the fit of each factor to each growth and phenological event with a step length of 2 to 60 (30 models in total), in which precipitation and sunshine length were the sums of multi-day values. However, other factors (average temperature, maximum temperature, minimum temperature, average soil temperature, relative humidity, and soil moisture) were represented by the average of multi-day values, and we used the measured values to indicate the value of 5 days before and after the measurement since the soil moisture was measured once every 10 days. There was collinearity between average and maximum temperature (Pearson´s correlation coefficient = 0.85), average temperature and minimum temperature (Pearson´s correlation coefficient = 0.85), average temperature and average soil temperature (Pearson´s correlation coefficient = 0.90), as well as maximum temperature and average soil temperature (Pearson´s correlation coefficient = 0.86) (Fig. S6). Three parallel models were constructed to remove collinearity, i.e., average temperature + precipitation + relative humidity+ soil moisture + sunshine length; maximum temperature+ minimum temperature + precipitation + relative humidity+ soil moisture + sunshine length; minimum temperature + soil temperature+ precipitation + relative humidity+ soil moisture + sunshine length, resulting in a total of 90 models (30 * 3 parallel models). Finally, we produced candidate models based on a cut-off of ΔAIC < 5 and subsequently selected the model with the shortest number of days as the best model among the candidate models. We calculated the average of the factors in multiple parallel models to select the best model, i.e., the multiple parallel models within the same day (Tables S3 and S4).
The ‘stats’ and ‘deSolve’ packages were employed for the logistic function, the ‘multcomp’ and ‘agricolae’ packages to perform variance analysis, and the ‘basicTrendline’ package for the linear model. We used the ‘PerformanceAnalytics’ package for correlation analysis and the ‘stats’ package for multiple regression analysis in R v. 3.5.1 (R Development Core Team).