Impact of elevated precipitation, nitrogen deposition and warming on soil respiration in a temperate desert

Soil respiration (R_s) is the most important source of carbon dioxide emissions from soil to atmosphere. However, it is unclear what the interactive response of R_s would be to environmental changes such as elevated precipitation, nitrogen (N) deposition and warming, especially in unique temperate desert ecosystems. To investigate this an in situ field experiment was conducted in the Gurbantunggut Desert, northwest China, from September 2014 to October 2016. The results showed that precipitation and N deposition significantly increased R_s, but warming decreased R_s, except in extreme precipitation events, which was mainly through its impact on the variation of soil moisture at 5 cm depth. In addition, the interactive response of R_s to combinations of the factors was much less than that of any single-factor, and the main response was a positive effect, except for the response from the interaction of increased precipitation and high N deposition (60 kg N ha⁻¹ yr⁻¹). Although R_s was found to show a unimodal change pattern with the variation of soil moisture, soil temperature and soil NH${}_{\mathrm{4}}^{+}$-N content, and it was significantly positively correlated to soil dissolved organic carbon (DOC) and pH, a structural equation model found that soil temperature was the most important controlling factor. Those results indicated that R_s was mainly interactively controlled by the soil multi-environmental factors and soil nutrients, and was very sensitive to elevated precipitation, N deposition and warming. However, the interactions of multiple factors largely reduced between-year variation of R_s more than any single-factor, suggesting that the carbon cycle in temperate deserts could be profoundly influenced by positive carbon–climate feedback.

1 Introduction

Global climate warming, changes in precipitation patterns and increased atmospheric nitrogen (N) deposition have all occurred since the Industrial Revolution, especially in temperate regions (IPCC, 2013), which will be expected to significantly change soil respiration (R_s), the most important source of carbon dioxide (CO₂) from soil to atmosphere (Wu et al., 2011): the annual CO₂ flux from R_s was 10-fold that of fossil fuel emissions (Eswaran et al., 1993; Batjes, 1996; Gougoulias et al., 2014). Therefore, even a small change in R_s will profoundly affect greenhouse gas balance and climate (Heimann and Reichstein, 2008). Although a number of experiments of the effects of warming, precipitation and N deposition on R_s have been conducted in alpine grasslands, tundra regions, peatlands and temperate forests (Lafleur and Humphreys, 2008; Strong et al., 2017; Yang et al., 2017; Zhao et al., 2017), studies in temperate desert ecosystems are scarce, especially those on the impact of these changes' interactions on R_s. A field study of multi-factor interactive effects on R_s was therefore conducted in a temperate desert ecosystem to help in understanding the response of R_s to climate change and N deposition in future and highlight the main driving factors.

R_s includes autotrophic respiration (R_A), which is mainly from plant roots and mycorrhizal activities, and heterotrophic respiration (R_H), which is mainly from the activities of microorganisms (Hanson et al., 2000). Soil moisture is a critical limiting factor for plant roots and microbial activities in desert ecosystems (Huang et al., 2015a): R_s significantly increased, by 47–70 %, in a degraded steppe in Inner Mongolia, China, due to increasing precipitation (Chen et al., 2013), with the effect especially strong in summer (Zhang et al., 2017). In addition, in arid ecosystems, increasing precipitation significantly stimulated plant growth, enhanced soil microbial activity and abundance (Huang et al., 2015a), and it changed soil nutrient and substrate concentration, such as dissolved organic carbon (DOC), inorganic nitrogen content, moisture and temperature (Huang et al., 2015b).

Warming significantly increased soil temperature, another important controlling factor for plant growth and microbial activity (Sheik et al., 2011; Huang et al., 2015a). R_s rates were significantly increased in a forest soil and Tibetan Plateau grassland by warming (Chen et al., 2017a), reducing R_s with decreasing soil moisture in the growing season, but increasing R_s in the non-growing season (Fang et al., 2017; Li et al., 2017); no significant impact was observed from warming (T. Liu et al., 2016). Therefore, how R_s is affected by warming induced variations in the soil environment is still unclear. In addition, low and short-term N deposition enhanced R_s, while higher and long-term N deposition inhibited R_s due to changes in plant growth and microbial activity (Zhu et al., 2017), but no impacts have also been reported (Luo et al., 2017; Zhang et al., 2017). A meta-analysis showed that the effects of N enrichment on soil CO₂ fluxes depended on temperature and soil properties (Zhong et al., 2016); desert soils may be even more sensitive to its variation.

A nation-wide analysis showed that warming, elevated N deposition and precipitation significantly increased R_s in China (Feng et al., 2017). Some studies have shown that the warming effect on R_s mainly depended on the variation of soil moisture in a dry forest soil (Li et al., 2017). Luo et al. (2008), using a modeling analysis, found that interactive effects became increasingly weaker with increasing intensity of the factors, but a recent meta-analysis showed that interactive effects were much greater than an individual factor (L. Y. Zhou et al., 2016). Thus how multi-factor interactions impact R_s is still unclear. Therefore, an in situ experiment was carried out in the Gurbantunggut Desert to (1) investigate the single-factor and interactive responses of R_s to warming, precipitation and N deposition and (2) identify the main controlling factors on R_s.

2 Materials and methods

2.1 Study site

A field experiment was carried out at the southern edge of the Gurbantunggut Desert (44^∘26^′ N, 87^∘54^′ E; 436.8 m a.s.l.), northwest China, from September 2014 to October 2016. This is the largest fixed/semi-fixed temperate desert in China. The mean annual temperature and precipitation are 7.1 ^∘C and 215.6 mm, respectively (Cui et al., 2017), and annual potential evaporation exceeds 2000 mm. From late November to mid-March, 20–35 cm of snow covers the whole desert (equivalent to 38–64 mm rainfall; Huang et al., 2015c). The growing season is from April to October. This desert soil is of extremely low fertility and is highly alkaline (Cui et al., 2017). Soil organic carbon, total N content, soil NO${}_{\mathrm{3}}^{-}$-N, NH${}_{\mathrm{4}}^{+}$-N contents and C : N ratio are 2.21 ± 0.71 g kg⁻¹, 0.08 ± 0.003 g kg⁻¹, 4.49 ± 0.71 mg kg⁻¹, 1.38 ± 0.74 mg kg⁻¹ and 21.39 ± 1.84, respectively (Table 1; Cui et al., 2017). Plant species are dominated by Haloxylon ammodendron and Haloxylon persicum, and vegetation is extremely sparse, with only 30 % coverage, with some spring ephemeral plants (May–June), annuals and perennial herbaceous plants (July–August; R. Liu et al., 2016). Spring ephemerals account for > 60 % of the community cover and 85 % of the biomass. Summer ephemerals, annuals and perennials usually account for only a small proportion of the community biomass before June, but dominate the community after the die-back of the spring annuals (Huang et al., 2015c).

Table 1 The annual, growing season (GS), non-growing season (NGS) and between-year fluxes and variation of soil respiration (R_s) over September 2014 to September 2016 (mean ± SE), including the contribution of GS and NGS, and the treatment effect. The positive values stand for increased R_s, and the negative value stand for reduced R_s. Dates are shown in YYYY.M format.

2.2 Experimental treatments

A striking N deposition rate (35.2 kg N ha⁻¹ yr⁻¹) exists in the Gurbantunggut Desert due to the rapid development of agriculture and industry, with the main form being ammonium nitrate (NH₄NO₃); deposition is split between wet (19.6 kg N ha⁻¹ yr⁻¹) and dry (15.6 kg N ha⁻¹ yr⁻¹; Song et al., 2015). In addition, according to the forecast of Galloway et al. (2008), atmospheric N deposition will double from the early 1990s to 2050; Liu et al. (2010) predict that precipitation in this region will increased by 30 % in next 30 years. In September 2014 to August 2016, an in situ complete block interactive experiment was therefore conducted to study the impact of N deposition and increased precipitation on R_s. The three levels of N deposition (0 kg N ha⁻¹ yr⁻¹, control, N0; 30 kg N ha⁻¹ yr⁻¹, low, N1; 60 kg N ha⁻¹ yr⁻¹, high, N2) and two levels of precipitation (“natural” precipitation, W0, and an increase of 30 % – an extra 60 mm precipitation annually, W1) were applied (Cui et al., 2017). Therefore there were six treatments (W0N0, W0N1, W0N2, W1N0, W1N1 and W1N2) with four replicates of each treatment. Each replicate plot was 10 m × 10 m with a 5 m wide buffer zone. The additional precipitation and N deposition (NH₄NO₃) were added 12 times in April, July and September, equivalent to 5 mm precipitation and 2.5 or 5 kg N ha⁻¹ per application over a week. The NH₄NO₃ was diluted in 50 L water (equal to 0.5 mm precipitation), and evenly applied following the simulated precipitation. The same amount of water was applied to the control plots (W0N0).

Figure 1 Comparative effects of warming from open-top chambers (OTCs), increased precipitation and N deposition on soil temperature (a), soil moisture (b) at 5 and 20 cm depth; soil organic carbon (SOC), NH${}_{\mathrm{4}}^{+}$-N content, NO${}_{\mathrm{3}}^{-}$-N content and pH (c), dissolved organic carbon (DOC), microbial biomass carbon (MBC) and microbial biomass nitrogen (MBN, d). The data are mean ± SE, n = 4 in (c) and (d); different letters indicate significant effect at P < 0.05.

Rapid warming (0.6 ^∘C per decade), increasing precipitation (3–5 mm yr⁻¹ since 1979) and high N deposition (3 kg N ha⁻¹ since 1980) are affecting the Gurbantunggut Desert (Liu et al., 2013; Li et al., 2015), which are excepted to affect rate of R_s. Therefore, another interactive experiment was established at the same time, simulating the three most likely climate scenarios in the future: (1) warming only (W0N0T1), (2) increased precipitation and N deposition without warming (W1N1T0), (3) the interaction of increasing precipitation, N deposition and warming (W1N1T1); all compared with the current climate (W0N0T0). Therefore, there were four treatments (W0N0T1, W1N1T0, W1N1T1, W0N0T0) with four replicates (plots) of each treatment. Open-top chambers (OTCs) were used to simulate warming. The OTCs were designed with 5 mm transparent tempered glass and stainless steel angle iron to the ITEX standard (Marion et al., 1997). They were 2 m high and 4 m in diameter, with each OTC area being 12 m². However, the design was improved such that the top and bottom OTC areas were the same so that precipitation and snowfall were the same as that in the surrounding environment; this also avoids overheating inside the OTCs. The timings of water and N applications were the same as in the complete block interactive experiment.

2.3 Measurements

R_s in all plots were measured twice or thrice a week (continuous measurements over 3 days were made following simulated precipitation and N deposition) using gas chromatography and static chambers (50 cm × 50 cm× 10 cm) at locations where only spring ephemeral plants grow, without any annuals and perennials, in order to minimize the between-treatment spatial heterogeneity due to sparse annuals and perennials (Liu et al., 2012). Gas samples were collected between 10:00 and 12:00 (GMT + 8) throughout the experimental period; measurements from this period were close to the diurnal averages (Figs. 3b, d and S1 in the Supplement). Gas samples were collected from the headspace of each chamber 0, 10, 20 and 30 min after closing the chamber. The gas samples were analyzed within 3 days using a gas chromatograph (GC; Agilent 7890A, Agilent Technologies, Santa Clara, CA, USA) equipped with a flame ionization detector for quantitative R_s (Liu et al., 2012). R_s rates were calculated from four concentrations of the gas samples based on a first-order differential linear or nonlinear equation and were temperature and pressure corrected (Liu et al., 2012; Zhang et al., 2014). Soil samples were taken monthly from around the static chambers to a depth of 10 cm using an auger (3.5 cm in diameter). Fine roots and small stones were separated out using a 2 mm sieve. Dissolved organic carbon (DOC) was extracted with deionized water (soil : water ratio = 1 : 10) by a centrifuge at 10000 rpm for 5 min and analyzed using a TOC analyzer (multi N/C 3100, Jena, Germany; Jones and Willett, 2006). The chloroform fumigation extraction of Brookes (1985) was used to measure microbial biomass carbon (MBC) and microbial biomass nitrogen (MBN). Soil organic carbon (SOC) was measured using the potassium dichromate method (Jiang et al., 2014), and soil NO${}_{\mathrm{3}}^{-}-$N and NH${}_{\mathrm{4}}^{+}$-N was analyzed as per Yue et al. (2016). Soil pH was measured with a 1 : 5 soil : deionized water suspension using a pH meter (Seven Easy, Mettler-Toledo, Switzerland). Caipos Soil and Environment Monitoring Systems (Caipos GmbH, Austria) were used to monitor soil moisture/temperature at 5 and 20 cm depth every hour.

Figure 2 Variation in rainfall (mm, a), and air temperature (^∘C, a) from September 2014 to October 2016 at the Gurbantunggut Desert. Also shown are soil moisture and temperature responses to increasing precipitation and warming (b), and the response of R_s (mean, n = 4) to precipitation (c), N deposition (d, e) and warming (f). W0 and W1 indicate under ambient precipitation (without water addition) and 60 mm yr⁻¹ precipitation addition; N0, N1 and N2 indicate 0, 30 and 60 kg N ha⁻¹ yr⁻¹ nitrogen addition; W1N0, W1N1 and W1N2 indicate 0, 30 and 60 kg N ha⁻¹ yr⁻¹ nitrogen addition under 60 mm yr⁻¹ precipitation addition; W1N1T1, W0N0T1 and W0N0T0 indicate the interaction between increasing precipitation (60 mm yr⁻¹), N deposition (30 kg N ha⁻¹ yr⁻¹) and warming by OTCs, warming alone (without increasing precipitation and N deposition) and control plots, respectively. Black arrows indicate simulated precipitation (5 mm) and N deposition (0.25 or 0.5 g N m⁻²). Each point represents the mean of four replications (chambers). Standard deviations for R_s are not shown for figure clarity.

2.4 Effects of each treatment on R_s

Each treatment effect was analyzed using the following formula to better evaluating the effect of precipitation, warming and N deposition on R_s (Yue et al., 2016).

$${\displaystyle }\text{Treatment effect}=(T{R}_{\mathrm{s}}-C{R}_{\mathrm{s}})/C{R}_{\mathrm{s}}\times \mathrm{100}\mathit{\%},$$

where the treatment effect is W0N1, W0N2, W1N0, W1N1, W1N2, W1N1T1 or W0N0T1 effect on R_s (a positive value shows that the treatment has increased R_s and a negative value shows decrease of R_s), corresponding R_T,s represents R_s from the W0N1, W0N2, W1N0, W1N1, W1N2, W1N1T1 or W0N0T1 plots (mg C m⁻² h⁻¹) and R_C,s indicates the R_s from the control plots (W0N0, mg C m⁻² h⁻¹).

2.5 Statistical analyses

Treatment effect on SOC, NO${}_{\mathrm{3}}^{-}$-N, NH${}_{\mathrm{4}}^{+}$-N content, pH, DOC, MBC and MBN were examined in each treatment by least significant difference (LSD; p < 0.05). The single-factor and interaction effects of precipitation, warming and N deposition on R_s were detected using multi-way analysis of variance (ANOVA), and the cumulative effect of precipitation, warming and N deposition on R_s were tested by repeated measures ANOVA. In addition, the relationships of R_s and DOC, MBC, MBN, soil temperature, soil moisture, NH${}_{\mathrm{4}}^{+}$-N content, soil NO${}_{\mathrm{3}}^{-}$-N and pH were described using a linear or nonlinear regression model. The factors of key controls on R_s were analyzed with structural equation models (SEMs). SPSS software (version 20.0) was used to conduct all statistical analyses, and statistically significant differences were set at P < 0.05. All figures were created using the Sigmaplot software package (version 10.0), but SEM analyses were carried out using AMOS 22.0 (Amos Development Corporation, Chicago, IL, USA).

3 Results

3.1 Treatment effects on soil environmental and properties

Soil temperatures at 5 cm depth mostly increased between 11:00 and 22:00 every day due to warming; the average annual soil temperatures at 5 and 20 cm depth increased significantly, by 4.41 and 3.67 ^∘C, respectively (Fig. 1a). Soil moisture at 5 cm depth was decreased by warming by only 0.61 v∕v % (Fig. 1b), and a very small decrease of 0.01 v∕v % in soil moisture at 20 cm depth was observed (Fig. 1b). Soil moisture at 5 and 20 cm depth were largely increased by the increased precipitation (Fig. 1b). N deposition and warming significant increased soil NH${}_{\mathrm{4}}^{+}$-N and NO${}_{\mathrm{3}}^{-}$-N contents (Fig. 1c), but no significant change was found from increased precipitation. Soil MBC and MBN were greatly increased by N deposition, but significant negative effects on soil MBC and MBN were observed by warming and the interaction of precipitation and N deposition (Fig. 1d). No significant change in SOC and DOC was observed in any treatment (Fig. 1c and d).

Figure 3 Post-extreme-rainfall diurnal variation in R_s (mean ± SE, n = 4, d) with variation in soil temperature and soil moisture (c), and extreme-drought variation in R_s (mean ± SE, n = 4, b) with variation in soil temperature (T5, T20, a) and soil moisture at 5 cm (W5) or 20 cm (W20) depth caused by warming in OTCs. Positive values indicate an increase due to warming, and negative values indicate decline. A red straight line indicates the average value of R_s inside the OTCs in (b) and (d), and a blue straight line represents the average value of R_s out of OTCs in (b) and (d). Red *, ** and *** symbols indicate significant effects at P < 0.05, P < 0.01 and P <0.001, respectively.

3.2 Precipitation, warming and N deposition effects on R_s

In our study, a weak R_s emission rate (−2.46 to 50.26 mg C m⁻² h⁻¹) was observed at control plots with an average emission rate of 12.18 mg C m⁻² h⁻¹ from September 2014 to October 2016 (Fig. 2c). Annual cumulative rate of R_s was 1090.11 ± 450.78 kg C ha⁻¹, with the non-growing season accounting for 20.7 % of the annual emission (Table 1). R_s was significantly enhanced by increasing 5 mm precipitation and N deposition from 12.18 to 16.23 and to 14.97 mg C m⁻² h⁻¹(average), respectively (P < 0.001; Fig. 2c and d; Table 2), with annual R_s increased by 33.1 % and 19.2–22.8 %, respectively (Table 1). The low N deposition effect on R_s was much higher than that of high N deposition (Fig. 2c and d). However, R_s was reduced mostly by warming, although not significant (P=0.084; Table 2). The high temperatures and low humidity at times of peak sunshine during the diurnal variation significantly inhibited the emission rate (Figs. 3a and b, S2), but it was also significantly increased by warming following extreme rainfall events that increased soil moisture (Fig. 3c and d). The diurnal trend in R_s was consistent with that of soil temperature at 5 cm depth (Fig. 3). In addition, the interactive responses of R_s to increasing precipitation, warming and N deposition were much lower than that from any single-factor (Table 1), and with the interaction of 60 kg ha⁻¹ N and extra precipitation decreasing R_s by 4.25 % (Table 1). Overall, annual R_s rates were significantly impacted by precipitation, N deposition, and their interaction (Table 2), but no significant net change was caused by warming (Table 2), although R_s rates were decreased by 9.99 % (Table 1).

Table 2 Tests of significance of year (Y), warming (T), precipitation (W) and nitrogen addition (N) on soil respiration (R_s) using multivariate ANOVA (F and P values). The cumulative effect of precipitation, N deposition and warming on R_s in 2015 and 2016 (F and P values) as assessed by repeated measures ANOVA. Superscripts a, b and c indicate significant effects at P < 0.05, 0.01 and 0.001, respectively.

3.3 Temporal variation and its control

The results of repeated measures ANOVA showed that significant cumulative effects on R_s were found to be due to N deposition and interaction between N deposition and precipitation or warming rather than increasing precipitation and warming alone (Table 2). A large between-year variation in R_s was observed with a coefficient of variation (CV) up to 41.4 % (a much higher R_s rate was observed in 2016 than 2015), but variation was reduced by increasing precipitation, N deposition and warming and their interaction, except with an increase in N deposition of 30 kg ha⁻¹ (Table 1). The results of regression analysis showed that R_s was significantly increased by increases in pH and DOC (Fig. 4e and f), but no significant relationships were found with MBC, MBN or NO${}_{\mathrm{3}}^{-}$-N content (Fig. 4a, b and c). In addition, different response characteristics of R_s in the impacts of increased soil moisture, soil temperature and NH${}_{\mathrm{4}}^{+}$-N content were found. Soil moisture was the most important controlling factor when it was < 4.2 % and soil temperature was > 26.5 ^∘C (Figs. 4g, h and S2). Secondly, soil temperature was the most important limiting factor when soil moisture was > 15.9 % and soil temperature < 2.7 ^∘C (Figs. 4g, h and S2). Thirdly, there was no significant impact on R_s when soil NH${}_{\mathrm{4}}^{+}$-N content was < 6.3 mg N kg⁻¹. A significant increase in R_s occurred when soil NH${}_{\mathrm{4}}^{+}$-N content was between 6.3 and 12.6 mg N kg⁻¹, but R_s was inhibited when soil NH${}_{\mathrm{4}}^{+}$-N content was between 12.6 and 31.6 mg N kg⁻¹ (Fig. 4d).

Figure 4 The relationship of soil respiration with microbial biomass carbon (MBC, a); microbial biomass carbon (MBN, b); soil NO₃-N (c); NH${}_{\mathrm{4}}^{+}$-N content (x, d); soil dissolved organic carbon (DOC, e); pH (f); soil moisture (g) and soil temperature (h).

4 Discussion

4.1 Single-factor impacts of precipitation, N deposition and warming on R_s

Annual R_s was 1090 kg C ha⁻¹ in this temperate desert, with the non-growing season accounting for 20.7 % of the annual flux (Table 1). This is consistent with previous studies here (Zhou et al., 2014; Huang et al., 2015a) because SOC content was very low (Fig. 1c), and vegetation was very sparse in this desert (R. Liu et al., 2016). Increasing precipitation significantly increased R_s (Fig. 2c). It is also consistent with the results of a meta-analysis and previous studies here (Huang et al., 2015a ; L. L. Liu et al., 2016). This is because the growth of desert plants and microbial activity are significantly activated by increasing precipitation (Huang et al., 2015a), and microbial biomass, mass-specific respiration, microbial biomass carbon (MBC) and nitrogen (MBN), and microbial PLFAs (phospholipid fatty acids) were consistently significantly enhanced by increased precipitation (Zhang et al., 2013; Huang et al., 2015a). However, R_s in our study was much higher in moderate soil moisture conditions than with too little or too much soil moisture (Fig. 4g). This suggests that R_s is mainly R_H rather than R_A in this desert, namely from soil microorganism, because (1) too little or too much soil moisture could significantly inhibit microbial activity due to variation of soil temperature and soil properties (Ma et al., 2013), while moderate soil moisture could significantly enhance microbial activity (Skopp et al., 1990) and (2) the biomass of fine roots was not significantly enhanced at our sites by increased precipitation (Cui et al., 2017). This is consistent with results from a desert steppe in northern China where the contribution of R_H (78.1 %) was significantly higher than that of R_A (21.9 %) under increasing precipitation (T. Liu et al., 2016).

N deposition also significantly increased R_s, especially in low N deposition (Fig. 2d). This is consistent with results from an alpine meadow and in the Loess Plateau (Fang et al., 2017; Zong et al., 2017), and with a meta-analysis showing that N deposition increased R_s by 8.8 % (L. Y. Zhou et al., 2016). This is because N deposition, on the one hand, could increase fine root biomass, although this was not significant in our study (Cui et al., 2017), and on the other hand, increases microbial activity and abundance by low N deposition (Huang et al., 2015b). However, this was inconsistent with a young Cunninghamia lanceolata forest (Wang et al., 2017), and beneath shrubs of H. ammodendron, soil high N content had the opposite effect as in our study site (Chen et al., 2013; Huang et al., 2015b). Additionally, the results of nonlinear regression analysis showed that higher R_s rates occurred at moderate soil NH${}_{\mathrm{4}}^{+}$-N contents (between 6.3 and 12.6 mg N kg⁻¹), while lower R_s occurred in much lower (< 6.3 mg N kg⁻¹) or much higher ( > 12.6 mg N kg⁻¹) soil NH${}_{\mathrm{4}}^{+}$-N contents (Fig. 4d); however, this effect of N deposition on R_s is not consistent with other ecosystems (Burton et al., 2004; Chen et al., 2013; Liu et al., 2015; Chen et al., 2017b). This is because the desert soil is extremely limited compared to other ecosystem (Adams, 2003), so low N deposition enhanced plant growth and microbial activity, but high N inhibited microbial activity and community composition, and reduced R_s (Zhou et al., 2014; Huang et al., 2015b). Overall, soil NH${}_{\mathrm{4}}^{+}$-N content was an important controlling factor for R_s because microbial activity, abundance and species diversity were regulated by soil NH${}_{\mathrm{4}}^{+}$-N content in this desert, and R_s was very sensitive to variation of N deposition.

Warming decreased R_s (Fig. 2f), although not significantly (P=0.084; Table 2), which was consistent with results from a semi-arid alfalfa pasture of the Loess Plateau (Fang et al., 2017). In addition, significant decreases in R_s were observed on extreme drought or hot sunny days when soil moisture was reduced, and sharply reduced R_s occurred when soil temperature reached 37 ^∘C (Figs. 3a, b and S1). This is because (i) microbial activity is significantly inhibited by extreme temperatures and low soil moisture may reduce population size by 50–80 % (Sheik et al., 2011) and (ii) fine root growth is inhibited in high temperature and low soil moisture. Others have noted this phenomenon as occurring at about 16:00 each day (Ma et al., 2013), but in our study the effect was advanced to 14:00 by warming, which may reduce carbon emission from soil to atmosphere. However, this is not consistent with results from a tundra ecosystem, subtropical forest or alpine regions where R_s was significantly increased by warming due to the limitation of soil temperature in these ecosystems, and not to significant change in soil moisture (Noh et al., 2016; Wu et al., 2016; Y. M. Zhou et al., 2016). In addition, a significant increase in R_s was found following enhanced precipitation with warming (Fig. 3c and d), which indicates that soil moisture was the most important controlling factor for R_s under a warming climate. This is consistent with other studies (Chen et al., 2017a; Zhao et al., 2017). However, statistical analysis showed that overall no significant impact on R_s was found during the experimental period due to warming (it was reduced by 9.99 %). This is because our gas samples were taken at 10:00–12:00 each day, when average soil temperatures were increased by about one degree. Thus mean annual R_s was not sensitive to temperature changes this small in contrast to the very significant effects of short-term diurnal changes in soil temperature observed between 12:00 and 17:00 (Fig. 3a and c). However, gas samples during 10:00–12:00 in this study was able to catch the mean warming effects on R_s, except in some extreme precipitation and drought events in summer (Fig. S1), which require further systematic evaluation. Those results indicated that R_s depends mainly on variations of soil moisture and temperature in the context of warming, and climate change is likely to have a very significant effect on temperate deserts.

Figure 5 Structure equation model (SEM) testing the multivariate (soil moisture, soil temperature, soil NH${}_{\mathrm{4}}^{+}$-N content, DOC and pH) effects on R_s (n = 34). Arrows show the effects of different key controls on R_s analyzed. The blue arrows indicated positive effects, and red arrows showed negative effects. The width of the arrows indicate the strength of the relationship. The numbers are standardized path coefficients, which can show the importance of the variables in the model. Goodness-of-fit statistics for the model are shown below the model. Asterisks indicate significant effect at P < 0.05.

4.2 The interactive effects of precipitation, N deposition and warming on R_s

Interactive responses of R_s were much lower than those of any single-factor, but still increased R_s overall, except with the interaction between precipitation and high N deposition (Table 1). This is consistent with results in dry ecosystems (Morillas et al., 2015; Martins et al., 2016), but not with the results of a meta-analysis of precipitation and N deposition interactive experiments which showed greater positive effects on R_s (L. Y. Zhou et al., 2016). This can be explained in our study due to soil MBC or MBN being much lower in interactive treatments than those of single-factor (Fig. 1d), showing that the number of microorganisms was much smaller in interactive treatments than that of single-factor due to the much stronger N effect. As we found that R_s was reduced with increasing N deposition and precipitation by as much as 4.25 % in W1N2 plots (Table 1), this showed that the inhibiting effect of soil NH${}_{\mathrm{4}}^{+}$-N content was much stronger when there was sufficient soil moisture (Fig. 2e). This is consistent with the results in a Populus euphratica community in a desert ecosystem (He et al., 2015). This was because (i) microbial activity was inhibited by high or low soil moisture and high soil NH${}_{\mathrm{4}}^{+}$-N or NO${}_{\mathrm{3}}^{-}$-N content (Burton et al., 2004) and (ii) high N content to reduce extracellular enzyme activity and the fungal population (Maris et al., 2015). In addition, the interactive effect of the three factors on R_s in this desert was much lower than interaction of two factors of precipitation and N deposition (Table 1), and is consistent with the results of modeled interactive effects, which showed that three-factor interactions were rare while two-factor interactions were more common (Luo et al., 2008). Fortunately, the interactive effect of three factors or two factors (precipitation, N deposition and warming) in this desert could largely reduce between-year variation on R_s (Table 2), which may because (i) the limits of soil moisture, soil temperature and soil N content were lifted for key biological processes by increasing precipitation, N deposition and warming (Huang et al., 2015a; R. Liu et al., 2016) and (ii) various factors counteracting each other (L. Y. Zhou et al., 2016). However, the effects of the variation on R_s in the growing season on R_s can be increased by warming, elevated precipitation and N deposition because of their dominant effects on plant growth and microbial activity (Huang et al., 2015b), but it was the exact opposite in the non-growing season due to the reduced temperature limitation (Zeng et al., 2016). Those results showed that R_s would be reduced under interactive effect of increasing rainfall, temperature and N deposition in the future, and positive carbon–climate feedback took place.

4.3 Temporal effects of the variation in treatments on R_s and controlling factors

Significant cumulative effects on R_s were found by elevated N deposition rather than increasing precipitation and warming alone (Table 2). A previous study here showed that R_s decreased due to N addition over the experiment's duration (Zhou et al., 2014), which was inconsistent with our results (Fig. 2d) because in our study, we used relatively lower rates of N addition than that Zhou et al. (2014), and the composition of microbial community and soil properties were altered gradually by long-term and high N deposition (Fig. 1c and d; Huang et al., 2015b; Zong et al., 2017). In addition, significant cumulative effects in the interaction between N deposition and precipitation or warming on R_s were also found (Table 2), and R_s decreased by 4.25 % by interaction between increasing precipitation and high N deposition (Table 1), which indicated that the response of R_s to N deposition is largely dependent on soil moisture in desert soil. This may be attributed to the antagonistic interaction between elevated N deposition and precipitation on R_h (L. Y. Zhou et al., 2016). Those results indicated that N deposition produced strong cumulative effects on R_s in this desert, and was enhanced largely with increasing soil moisture, which would reduce carbon emission from soil to atmosphere.

Regression analysis shows that R_s exhibited a unimodal change pattern with variations of soil NH${}_{\mathrm{4}}^{+}$-N (Fig. 4d), moisture (Fig. 4g), and temperature (Fig. 4h), and R_s was significantly positively correlated to soil dissolved organic carbon (DOC) and pH (Fig. 4e and f). However, structural equation modeling indicated that soil temperature was a more important controlling factor than soil NH${}_{\mathrm{4}}^{+}$-N and soil moisture (Fig. 5). This is consistent with most research results (Wu et al., 2016; Y. M. Zhou et al., 2016; Chen et al., 2017a). In addition, a large interannual variation was observed (CV = 41.4 %) during our experiment (Table 1), while the variation of annual precipitation and air temperature were only 4.41 and 7.78 %, respectively (Table 1), but close to the CV of spring root biomass of ephemeral plants (47.14 %), with 24 times greater aboveground biomass of spring ephemeral plants in 2016 than in 2015 (Cui et al., 2017), which indicated that the increase of R_s in 2016 was mainly from the root respiration of ephemeral plants. This is consistent with previous studies showing that ephemeral plants mediated the interannual variations of carbon fluxes in this desert (Huang et al., 2015c; R. Liu et al., 2016). It is different from other ecosystems where interannual variations of R_s were mainly dependent on variations in annual precipitation and air temperature (Gerard et al., 1999; Asensio et al., 2007; Chen et al., 2012). Overall, our results indicate that annual variation in R_s in this temperate desert is mainly controlled by soil temperature, but between-year variation in R_s is mainly controlled by ephemeral plants.

5 Conclusions

Climate change and elevated N deposition play important roles in controlling R_s in temperate deserts. We found that increasing precipitation and N deposition significantly increased R_s in the Gurbantunggut Desert, but warming reduced R_s, mostly because of the variation of soil moisture. In addition, we found that the interactive responses of R_s were much lower for the combination of precipitation, N deposition and warming than for any single factor. Additionally, R_s is mainly mediated by soil moisture, soil temperature and soil NH${}_{\mathrm{4}}^{+}$-N content, but soil temperature is the most important, with between-year variation in R_s mainly controlled by ephemeral plants. These results showed that R_s is very sensitive to increasing precipitation, N deposition and warming, and their interactive effects could reduce soil carbon emissions and thus the impacts of climate change.