Study site
We conducted this study at the Maodeng Grassland Ecosystem Research Station of Inner Mongolia University (44°10’ N, 116°28’ E, 1101 m a.s.l.) located in the Xilingol region of Inner Mongolia, China (Figure S1a). This area has a temperate semi-arid climate, with a short and cool growing-season that normally starts in May and ends in October. During the 3 years of study (2017–2019), the mean growing-season and annual air temperature was 10.7 °C and –1.2 °C, respectively. Annual growing-season precipitation ranged from 137 to 252 mm yr–1 and accounted for ~90% of total annual precipitation (ranged from 168–278 mm, Figure S2). The natural vegetation is a typical steppe dominated by perennial grasses such as Leymus chinensis, Stipa baicalensis and Cleistogenes squarrosa. Soil developed is a Calcic-Orthic Aridisol according to the US soil taxonomy classification system with a mean pH of 8.1 and a mean soil organic carbon content of 12.5 g kg–1 at 0–20 cm soil depth (Wang et al. 2020b).
Experimental design
We established a field experiment with treatments of precipitation amount and land-use regime using a randomized complete block design with split plot (Figure S1b). Within each block, we arranged land use treatments in plots and precipitation amount treatments in subplots. Specifically, in 2011, we established three 100 m × 100 m blocks with 3 m in between. Each block was divided into nine 33.3 m × 33.3 m plots. Within each block, we randomly assigned three land use treatments to 3 plots. The three land use treatments are no grazing or mowing (wire fence enclosure), grazing (with 6 sheep in July and August, once per month until residual height of plants reached 6 cm), and mowing (to 6 cm, once per year in August). The intensity of grazing and mowing treatments in our experiment represented a moderate and commonly practiced land-use intensity in this region (Baoyin et al. 2014). In 2016, we established precipitation amount treatments at the end of the growing season. Briefly, we delineated three 3 m × 5 m subplots with 2 m in between within each land use treatment plot. Each subplot was randomly assigned to one of the three precipitation treatments. Around each subplot, we inserted metal plates to a depth of 15 cm with 5 cm remaining aboveground to prevent lateral move of rainfall. Consistent with a network of world-wide precipitation manipulation experiments (www.drought-net.org), we implemented three levels of precipitation treatment: 50% reduction (Dry), ambient (Amb), and 50% increase (Wet). Reduction in precipitation was achieved by installing ten transparent sheet channels over the subplots of dry treatment (25 cm wide and 340 cm long). These channels were made of transparent Panlite (Zhuonier Corporate, Suzhou, China) with a high light transmission ratio (>90%) to minimize shading. The channels were installed at a ~10° angle and covered 50% of subplot so that 50% of rainwater were intercepted (Figure S1b). The intercepted rainwater was collected and immediately sprinkled to a wet treatment subplot within the same plot after the rain event, resulting in a 50% increase in precipitation for the wet treatment subplot. During the 3 years of study (2017–2019), growing-season precipitation ranged from 137–252 mm/yr. The range of precipitation received by all treatments (68–378 mm/yr) represents 75% of the growing-season precipitation range observed in the last 10 years (136–454 mm/yr) (Figure S3).
Soil respiration measurement
We manually measured Rs twice a month during the growing season (from May to October, Figure S4–S8). Measurements were made between 8:00 AM–12:00 PM on a sunny day of the 1st and 3rd weeks of each month using a LI-8100 Automated Soil CO2 Flux systems (Li-Cor Inc., Lincoln, NE, USA) on a shallow polyvinyl chloride (PVC) collar (20 cm in diameter and 10 cm in height) installed to a soil depth of 5 cm. We removed all above-ground vegetation inside the collar. We also measured respiration over a deep PVC collar (20 cm in diameter and 45 cm in height) installed to a soil depth of 40 cm with all above-ground vegetation inside the collar removed. These measurements represent the heterotrophic respiration because more than 80% of the belowground biomass is distributed in the top 30cm and the installation depth is sufficient to exclude most organic matter input from adjacent plants (Figure S9). We installed these deep collars 9 months prior to Rh measurements to eliminate experimental artifacts arising from the pulse of dead root input after collar installation (Zhou et al. 2007; Wang et al. 2021). Ra was calculated as the difference between Rs and Rh. We are fully aware of the potential artifacts associated with the collar methods for soil respiration measurements, such as removing vegetation on soil moisture and temperature inside the collar (Hanson et al. 2000; Kuzyakov and Larionova 2005; Baggs 2006). However, these effects are likely similar across treatments and thus have minimal influence in analyzing the treatment effects.
Concurrent with soil respiration measurements, we measured soil temperature (ST) and moisture (SM) at 5 cm depth using 6000-09TC and GS-1 probes attached to the LI-8100. Such a soil depth is commonly used to measure ST and SM in previous studies (Feng et al. 2018; Wang et al. 2018, 2021) because temperature measured at this depth usually had the highest explanatory power on Rs (Phillips et al. 2011; Wang et al. 2014). We also obtained daily air temperature and precipitation from a nearby weather station (~100 m).
Plant and soil sampling
In early August of each study year, we measured the aboveground (ANPP) and belowground (BNPP) net primary productivities. We used a harvest method for ANPP measurements (Xu et al. 2015b; Liu et al. 2018). Specifically, we prepared three 1 m × 1 m quadrats in each subplot and randomly selected one for each growing season. The selected quadrates within grazing plots were protected with cages, representing a long-term effect of grazing on ANPP (Milchunas and Lauenroth 1993; McNaughton et al. 1996). In early August, all green plant tissues within the selected quadrats were harvested, oven dried and weighed to obtain ANPP. We used a root ingrowth-core method (Liu et al. 2018) instead of a harvest method to measure BNPP because the plant community in this region is dominated by perennial species (Bai et al. 2004; Zhang et al. 2017; Wang et al. 2020c) and belowground biomass represents multi-year accumulation of root biomass production. We first took three soil cores (7 cm in diameter) of 0–50 cm along the diagonal of each subplot at the end of growing season in 2017, sieved (2 mm mesh) soils to remove roots and refilled soil cores with root-free soils collected from the same depth. During the following two years (2018 and 2019), we resampled soils from the same cores at the same time of ANPP measurement, sieved (2 mm mesh) soil samples to obtain roots, oven dried and weighed them to calculate BNPP. The residual root-free soils were put back for BNPP measurements the next year.
In early August of 2019, we sampled three additional soil cores along the diagonal of each subplot and measured microbial biomass carbon (MBC) and nitrogen (MBN) using the chloroform fumigation-extraction method (Vance et al. 1987). Specifically, we mixed three cores of soils collected from the same subplot, sieved through a 2 mm mesh to remove plant tissues, and then weighed 6 aliquots (3 g equivalent). Three of them were fumigated with ethanol-free CHCl3 at 25 °C in darkness for 48 h and the other 3 aliquots were unfumigated. The fumigated and unfumigated samples were extracted with 0.5 M K2SO4 (12 ml) for 30 min on a shaker. Carbon and nitrogen in the K2SO4 extracts were analyzed with a total organic C/N analyzer (Elementar vario TOC, Elementar Co., Germany) and the differences between fumigated and unfumigated samples were converted to MBC and MBN with a conversion factor of 0.45 (Brookes et al. 1985).
Statistical analysis
In our statistical analyses, the data of ST, SM, ANPP, BNPP, MBC and MBN were directly analyzed, while the data of Rs and its components (Rh and Ra) were first log-transformed to conform to the assumption of normal residuals (Lloyd and Taylor 1994; Yvon-Durocher et al. 2012) and then analyzed. We analyzed the effects of precipitation and land-use treatments on ST, SM, Rs and its components that are measured twice a month using repeated-measure analyses of variance (RMANOVAs). To explore the average effect over the three years of study and potential interannual variations, we performed this analysis separately in each year and using all three years’ data combined. Similarly, we analyzed the treatment effect of precipitation and land use on ANPP and BNPP using data from each year separately or three years combined. Because ANPP and BNPP were measured once per year, we used two-way ANOVAs when analyzing data from each year and RMANOVAs when data from all three years are combined. Finally, we analyzed the treatment effects on MBC and MBN using two-way ANOVAs because MBC and MBN were only measured once during the experiment period. In these analyses, treatments of precipitation and land-use regimes, and date (or year for ANPP and BNPP) of measurement were treated as categorical fixed effects, and block, plot and subplot were hierarchically arranged as random effects.
We used structural equation models (SEMs) to quantify direct and indirect impacts of abiotic (ST and SM) and biotic (plant and soil microbial parameters) factors on autotrophic (Ra) and heterotrophic (Rh) respirations. We constructed initial SEMs including all potential paths based on theoretical knowledge (Liu et al. 2009; Geng et al. 2012; Xu et al. 2015a; Chen et al. 2016) of drivers of Ra and Rh. In addition, these driving effects were also visually examined with linear regressions, such as abiotic (ST and SM) and biotic (plant and soil microbial parameters) factors in impacting Rs and its components, as well as abiotic factors in affecting biotic factors (see Figure S10–S12 for details). Subsequently, we performed a model selection of SEMs with the Chi-square ( ) difference test, the goodness-of-fit index (GFI) and the root mean-square error of approximation (RMSEA) to obtain final SEMs that have a good model fit (a small with P > 0.1, GFI > 0.90 and RMSEA < 0.10) (Kline 2005). In SEMs, we used the annual mean ST, SM, Rh and Ra in 2019 (27 observations for each variable) because MBC and MBN were measured only once in this year. We constructed separate SEMs for Ra and Rh, because Ra is calculated as the difference between Rs and Rh and thus is not independent of Rh. In these SEMs, all variables were standardized for ease of interpretation and comparison.
All statistical analyses were conducted using R 3.6.3 (R Core Team 2013). We used lavaan package for SEMs. All figures were produced using ggplot2 package. Effects were considered statistically significant if P < 0.05 or marginally significant if 0.05 ≤ P < 0.10.