Experimental site
This study was conducted in the Urat Desert‐grassland Ecosystem Research Station (106°58′E, 41°25′N, 1,650 m above sea level) located in western Inner Mongolia, China. The region has a temperate continental monsoon climate, and the mean annual precipitation is 139.5 mm, about 70% occurring during the growing season [63]. Main soil type in the study area is brown calcium, and the dominant species in the desert steppe are Stipa glareosa, Peganum harmala and Allium polyrhizum (Additional file 1 Table S1).
Experimental treatments
The extreme drought experiment was established in 2014 and was conducted from 2015 to 2018. This experiment involved three treatments: (1) a control (ambient precipitation, without shelters), (2) a -66% drought treatment (66% reduction from May 1 to August 31, with shelters), (3) and a -60 Days drought treatment (100% reduction from June 1 to July 31, with shelters). There are eighteen 6 × 6 m plots in total, which are randomly distributed in location and organized into six blocks. Each plot was located at least 2 m from the nearest neighboring plot and established a 1-m external buffer to minimize the edge effects. To prevent hydrological exchange with the surrounding soil, a 1 m deep sheet of plastic flashing was established in each plot. The roofs consisted of strips of clear polycarbonate plastic was situated 2 m above the ground at the highest point, which allowed for the circulation of air and avoided microclimatic changes. Polycarbonate plastic has been confirmed to have minimal influence on photosynthetically active radiation [64].
Sampling and analysis
During the peak of each growing season from 2015-2018, a quadrat (1 × 1 m) was set up in each experimental plot for vegetation investigation and sampling. Quadrat was marked to prevent subsequent resampling in the next year. We measured the number and the maximum height of each species within quadrat. Besides, we harvested all aboveground biomass (AGB) by species in each quadrat. Finally, we estimated belowground biomass (BGB) using a root auger (8 cm diameter) to measure root mass at a depth of 0-20 cm. The roots samples were taken back to the laboratory and then were washed free of soil over a mesh sieve (mesh size of 0.25 mm). All above- and belowground biomasses were dried at 65 °C in an oven for 48 h and weighed in the lab.
We determined five key functional traits to reflect the plant morphology and growth investment [65, 66]: plant height, specific leaf area (SLA), leaf dry matter content (LDMC), leaf carbon content (LCC) and leaf nitrogen content (LNC). These traits were measured for the dominant species making up 90% of the total plant cover in each plot. The five traits on 10 individuals per species in each plot were obtained by using the standard methodologies [67]. We calculated community-weighted means (CWM) of single-trait by multiplying the trait value of each species by its relative biomass in the community [68]. CWM can reflect the characteristics of community functional traits [69]. In each plot, three soil samples (0–10 cm depth) were collected to determine soil water, and one mixed soil sample from three random replicates was collected to measure soil organic carbon and total nitrogen content. Leaf carbon and nitrogen content (%), as well as soil organic carbon and total nitrogen content (g Kg-1), were measured by using an Elemental Analyzer [24] (Costech ECS 4010, Italy) with a reduction temperature of 650°C and a combustion temperature of 980°C.
Data analysis
The annual precipitation gradually decreased from 2015 to 2017, and increased in 2018 due to increased precipitation in July and August (Additional file 1 Figure S1). The precipitation in the early time of 2015-2018 growing season (March to June) was 56.2mm, 77.9mm, 28.8mm and 21.4 mm, respectively (Additional file 1 Figure S1), which decreased in an inter-annual timescale. Thus, we assess how experimental drought and natural drought in the early period of growing season affected structure and function of grassland ecosystems.
We analyzed the response of each variable to extreme drought using separate repeated measures mixed model ANOVAs with year, treatment, and their interaction as fixed factor and block as a random factor (Additional file 1 Table S4). One-way ANOVA was conducted to assess the significant differences of species richness, Density, AGB, BGB, CWM of Height, CWM of SLA, CWM of LDMC, CWM of LCC, CWM of LNC, Soil Carbon, Soil Nitrogen, and Soil Water over to extreme drought among years. A level of P < 0.05 was considered significant. Data are presented as mean ± standard error throughout.
Then, the simple regression models with a standard 95% confidence range were used to assess whether CWM of traits and soil factors could explain AGB and BGB. Based on the simple regression and the Correlation coefficients of each variable (Additional file 1 Table S3), a structural equation modeling (SEM) was performed, in which drought treatment and the precipitation in the early time were treated as exogenous variables; species diversity, CWM of trait, and soil factors were considered as endogenous variables; AGB and BGB were regarded as the response variable. We assessed the best fitting model using a Chi-square test, root mean square error of approximation and goodness-of-fit index [19], which was performed by AMOS 20.0 (Amos Development, Spring House, PA, USA).
Data analysis and plotting were run with the SPSS16.0 and SigmaPlot12.0 for Windows statistics program, respectively. The simple regression models were performed using the trendline function in the basic Trendline package of R software (v4.0.0, R Core Team, 2020).