Soil samples were collected from Daban Mountain (37°20′16.93″N, 101°23′47.21″E, 3,705 m a.s.l.), located in the eastern part of the Qilian Mountains in Qinghai Province. The area has a typical plateau continental climate, with a long cold winter and short warm summer. The annual average temperature was −1.6°C, with the maximum monthly mean temperature in July (10.1°C) and the minimum monthly mean temperature in January (−15.0°C). The historical extreme maximum and minimum temperatures were 26.8 and −37.1°C, respectively. The number of days with a daily minimum temperature below 0°C during the year was as high as 280 days. The annual precipitation was 560 mm, on average, of which 85% was concentrated in May–September. The regional annual mean evaporation was 1,238.0 mm (Li et al., 2004; Dai et al., 2019). The dominant species is Kobresia pygmaea, and it is associated with Saussurea, Potentilla, Leontopodium, Gentian, Saxifraga, Poa, Oxytropis, and Polygonum. The average height of vegetation was < 10 cm, and the aboveground dry biomass averages 210 g m^–2.

Soil samples were collected in a transition zone of seasonal permafrost and spot (island) permafrost, and the soil was classified as Inceptisol or Cambisol. The topsoil (0–10 cm) organic matter, bulk density, pH, and soil volume water content in K. humilis meadow are 138.52 ± 13.82 g kg^–1, 0.75 ± 0.05 g cm^–3, 7.50 ± 0.22, and 32.7% ± 5.17%, respectively (Wang et al., 2011). The site froze from late October to mid-November. A stable thin permafrost layer began to form in late November, and the thickness of the permafrost continued to increase and reached the maximum freezing depth of 0.5–1 m in mid-February of the following year. The soil began to enter the thawing period in early March, and the thaw depth continued to increase until thawing was complete by late April (Lin et al., 2017; Dai et al., 2019).

We used soil VWC to express the level of soil moisture. Three VWC values (W¹⁵–15%, W³⁰–30%, and W⁴⁵–45%) and three thawing temperature (T⁵–5°C, T¹⁰–10°C, and T²⁰–20°C) levels were applied in a randomized complete block design, for nine total treatments and three replications for each treatment. The topsoil of the study site usually begins to thaw from early-March to mid-April (Lin et al., 2017; Dai et al., 2019), and during this time, the day air temperatures fluctuate between 0°C and 13°C. Therefore, the 5°C of thawing temperature in our study was used as the control temperature. The 10°C of thawing temperature could represent the century-scale warming trend (2–7°C) of the QTP (Qin, 2014). The 20°C means extreme high temperature. According to our pre-experiment, the maximum volumetric water content was about 45%, and field study showed that the soil moisture in the growing season (July–August) were maintained at about 25%–35% VWC. Thus, we take the 30% VWC as the medium soil water. The difference in moisture gradient was controlled by 15% VWC. The existing precipitation control test also set treatments with natural and 50% reduction or increase in rainfall in an alpine grassland on the northeastern Tibetan Plateau (Xu et al., 2017).

In early July 2019, we set three 1-m × 1-m plots in the experiment site, then cut off the aboveground biomass, collected a sufficient amount of top soil (0–20 cm) with a drill (diameter 10 cm), and brought soil back to the laboratory the same day. After removing dead grass litter, plant roots, and gravel (grain diameter > 2 mm), 500 g of soil was weighed and transferred to a 1-L glass jar; then, the bottle was shaken to ensure that the soil contacted the bottle. The height of the soil in the bottle was 8–10 cm, and the soil surface was 11–13 cm away from the bottle plug. The plastic plugs of each bottle had three holes, two of which were connected to a rubber tube and a three-way valve. One of the three-way valves was linked with sampling needle tubing (10 ml), and the other valve was connected with the needle tubing (100 ml) to balance the bottle pressure. The third hole was used to hold the probe thermometer for recording the air temperature in each bottle. The interfaces of the holes were treated with latex to ensure airtightness. A time-domain reflectometer was used to measure soil moisture, which was adjusted in every treatment to the required VWC by adding ultrapure water. Each bottle was measured every 5 days during the whole test, and according to the change in mass, ultrapure water was added to keep the soil moisture stable.

Because the daily maximum temperature in the study area was nearly 15°C when we collected the samples, in order to make the test soil microbial community as consistent as possible with the early cold season, all samples were successively incubated at 15°C, 10°C, and 5°C for 1 week. After the cold acclimation, samples were put in three different incubators, and day and night F–T treatment was initiated. The freezing temperature was set as −20°C from 20:00 to 8:00 the next day, and thawing temperatures were set as 5°C, 10°C, and 20°C from 08:00 to 20:00. On the day of sampling, the freezing time were reduced by 90 min (21:30 to 08:00 the next day). The F–T cycles lasted about 165 days.

Gas collection and measurement were similar to the static chamber-gas chromatography method. Gasses were sampled once a day during the first 7 days, then once every 3 days from the 8th to 22nd day, once every 5 days from the 23rd to 48th day, and once every 7 days from the 49th to 165th days. The gas collection began at 20:00, and after the bottle was closed, a 10-ml gas sample was taken from the bottle with a plastic syringe at 0, 30, 60, and 90 min. Every time before the gas sample was drawn, the syringe piston was quickly pushed and pulled twice to stir and mix the air in the bottle. The three-way valves of the air intake channel were kept closed between the sampling intervals. We recorded the air temperature inside and outside of the bottle at every sampling, which were used to correct the N₂O flux. The soil surface temperatures of each treatment were measured with an infrared detector.

The N₂O concentration of the gas sample was analyzed within 24 h using gas chromatography (Agilent 7890A, Agilent Technologies, Sta Clara, CA, United States). The N₂O flux was calculated according to the following equation:

where F (μg m^–2 h^–1) is the N₂O flux; dc/dt is the slope of the linear regression for the N₂O concentration gradient as a function of time; M (g mol^–1) is the molecular mass of N₂O; P (Pa) is the atmospheric pressure; T (K) is the absolute temperature during sampling; V₀ (L), T₀ (K), and P₀ (Pa) are the gas mole volume, absolute air temperature, and atmospheric pressure under standard condition, respectively; and H (m) is the height of the soil surface to the bottle plug (Chen et al., 2018).

All data were assessed for normality of variance before analysis. First, the effects of VWC, temperature, number of F–T cycles, and their interactions on mean N₂O flux were investigated using general linear-model multivariate analysis. In the model, the significant differences in the mean N₂O fluxes between the treatments based on a least significant difference (LSD) test (IBM SPSS Statistics 20.0, SPSS Inc., Chicago, IL, United States). Second, a quarter of the measured N₂O flux was used to represent the daily mean emission rate, and the cumulative emissions were estimated using a linear interpolation method. Finally, linear, polynomial, exponential, and peak functions were used for fitting N₂O flux and soil moisture and temperature in OriginPro 2020b (OriginLab Corp., Northampton, MA, United States). The models with the highest fitting degree were selected to represent the regression relationship between N₂O flux and other environmental factors. All significant differences were at 0.05 (α).

The N₂O emissions of W³⁰ treatment were higher than W¹⁵ and W⁴⁵, and its patterns were similar even at different thawing temperatures. Figures 1A-1,B-1,C-1 shows that the emission periods of W³⁰ treatment were concentrated in the first 48 cycles, and the peak value occurred at the 13th F–T cycle. The peaks of W³⁰T⁵, W³⁰T¹⁰, and W³⁰T²⁰ were, respectively, 5,849.54, 1,123.16, and 1,253.83 μg m^–2 h^–1, which were 1,466.05, 335.27, and 327.37 times higher than the initial values. However, the major emissions of the W¹ group occurred from 1 to 27 days, the emission peak (190.78–223.27 μg m^–2 h^–1) occurred after the third or fourth F–T cycle and was only 1.55–1.85 times that at the beginning of the F–T cycles. Instead of increasing, the fluxes of W⁴⁵ treatments dropped sharply after the first cycle and were maintained at between −2.95 and 13.69 μg m^–2 h^–1 during all F–T cycles.

Variance analysis also indicated that at the same thawing temperature, the average fluxes of W³⁰ treatment were significantly higher than those of W¹⁵, and the W⁴⁵ treatment had the lowest fluxes (Figures 1A-2,B-2,C-2). At the same moisture, the average fluxes were not significantly different between different thawing temperatures in the W¹⁵ and W⁴⁵ groups, but the mean flux of W³⁰T⁵ (747.02 μg m^–2 h^–1) was significantly higher than that of W³⁰T²⁰ (363.99 μg m^–2 h^–1), and W³⁰T¹⁰ (220.41 μg m^–2 h^–1) was the lowest.

The three soil moisture levels showed three cumulative emission patterns. The total cumulative emissions of the W³⁰ group were the highest (102.04–366.6 mg m^–2), followed by the W¹⁵ group (24.96–34.18 mg m^–2) and the W⁴⁵ group (0.81–2.96 mg⋅m^–2). In the W¹⁵ group, the N₂O cumulative emissions continued increasing over all F–T cycles (average accumulation rate, 0.19 mg m^–2 day^–1) (Figures 2A–C). The accumulations of the W³⁰ group were mainly reflected at the beginning of F–T cycles (average accumulation rate, 8.3 mg m^–2 day^–1) and increased minimally after 60 cycles (Figures 2D–F). However, N₂O fluxes of the W⁴⁵ group were low in early F–T cycles, only increasing after 80 cycles, primarily in W⁴⁵ (Figures 2G–I).

During the early freeze–thaw cycle, ammonium increased significantly in all treatments, and the peak value range from 57.42 to 188.80 mg kg^–1 (Figures 3A-1,B-1,C-1). The increase in soil ammonium contents coincided with N₂O large emission in W¹⁵ and W³⁰. Although the peak ammonium contents in the water-saturated soil were higher than other treatments at the early freeze–thaw stage, there was no significant N₂O emissions. The nitrate content increased sharply in W³⁰T⁵ during early freeze–thaw cycle, and the peak value of nitrate content was nearly 24.11 mg kg^–1 (Figure 3A-2). Meanwhile, only low thawing temperature (T⁵) improved nitrate accumulation after long freeze–thaw cycles (Figures 3A-2,B-2,C-2).

The results showed that soil moisture, thawing temperature, and the number of F–T cycles all had significant impacts on N₂O flux, and their interactions were significant (P < 0.01) (Table 1). Figure 4 shows that the flux response to soil moisture during F-T cycles conformed to the Gauss Amp function at the three different thawing temperatures (P < 0.01). In the range of VWC 15–45%, N₂O flux variations were “unimodal.” The peak emission values of T⁵, T¹⁰, and T²⁰ treatments were at VWC 27%, 25%, and 30%, respectively, and the 95% confidence intervals were, respectively, 22–32%, 17–32%, and 22–38% (Figure 4). These results highlighted that the medium level of soil water was essential in controlling large N₂O emissions during F–T cycles.

There were no correlations between the N₂O flux and thawing temperature in the W¹⁵ and W⁴⁵ groups (Figures 5A,C). The response of flux to temperature showed a polynomial relationship (P = 0.02) in the W³⁰ treatment, and it increased with the temperature increase in the range of 0°C to 5°C, then decreased in the range of 5–10°C and increased again from 10°C to 20°C (Figure 5B). This suggested that the N₂O flux was mostly influenced by a medium moisture level in soil, and the large N₂O emissions during F–T cycles had complicated nonlinear relationships with the thawing temperature.

Based on the relationship between soil N₂O dynamics and soil water and thawing temperature in the first 48 F–T cycles (large emission period), we drew the diagram of N₂O flux model (Figure 6). We speculated that the medium moisture controlled the large production and release of N₂O over F–T cycles. Because of appropriate soil, porosity tends to result in abundant aerobic and anaerobic microenvironments. These microzones were especially common at low thawing temperatures when the ice crystals in deep soil do not melt completely and coexist with soil air and free water, simultaneously promoting nitrification and denitrification. However, under dry conditions, the vast aerobic environment was beneficial to nitrification, and some N₂O was released as the nitrification by-product. For the high soil water content, N₂O production was dominated by denitrification, and the large intermediate products of N₂O were completely converted to N₂ under an anaerobic environment. Moreover, under high moisture, the large interstices of deep soil were filled with water or ice, and the upper layer was covered with water. These physical barriers would block N₂O emissions and greatly reduce N₂O flux when frozen soil was thawing. The effect of thawing temperatures on the redox conditions of dry and waterlogged soil was limited, so the response of soil N₂O flux to thawing temperature tended to be consistent in these two states.

We found that no matter how the thawing temperature changed, the N₂O fluxes had significant peak functions with soil VWC, and the 95% confidence intervals ranged from 17 to 38% (VWC). These findings show that the relationship between soil moisture and N₂O flux during F–T cycles was not linear, and a medium level of moisture is more likely to promote N₂O emissions. In contrast, when the soil was dry or waterlogged, N₂O emissions were sharply reduced during F–T. Our previous in situ test of seasonal frozen arable soils in northeast China also indicated that soil moisture explained 32% of the N₂O flux variation, and high soil moisture (30–55%) triggered the N₂O burst during the spring thaw (Chen et al., 2018). Quick warming of the bare soil after snow melt resulted in rapidly decreased water content that returned to normal along with N₂O flux (Chen et al., 2018). However, in our earlier research, the soil moisture (55%) at maximum emissions was higher than in this study (17%–38%), which may be related to the high water-holding capacity of arable black soil (mollisols) (Chen et al., 2018).

Teepe et al. (2004) explained that N₂O emissions during thawing decreased in the order 64% > 55% > 42% [water-filled pore space (WFPS)], but the flux at 76% WFPS was less than that at 55% WFPS. If our soil moisture were converted into WFPS, the peak values corresponding to the regression curve were 61% (T⁵), 55% (T¹⁰), and 66% (T²⁰). This means that the flux was positively correlated with the soil moisture when WFPS was lower than 55%, while it was negatively correlated with moisture when WFPS was higher than 66%. Thus, we assumed that the soil moisture had a threshold value to determine the large N₂O release during F–T. Other laboratory experiments also support our inference. For instance, Lu et al. (2015b, a) studied meadow-steppe grassland soil and found that N₂O emissions in F–T processes are 30 μg kg^–1 h^–1 when soil moisture was about 8% VWC; however, the flux plunged to 8.4 μg kg^–1 h^–1 in waterlogged soil. If soil water is a crucial factor that controls N₂O emissions during F–T, it is easy to understand why the current results of global N₂O variation are so large (0.7–27.2 kg N ha^–1) (Chen Z. et al., 2016). If more informed studies of in situ data of N₂O fluxes and soil moisture in the F–T period were integrated by meta-analysis, more meaningful and accurate conclusions may be obtained.

We suspected that the effects of soil moisture on N₂O emissions during F–T cycles were largely related to nitrification and denitrification, both of which generate N₂O and are regulated by soil O₂ partial pressure. Studies show that, in dry soil, the N₂O emissions of nitrification increase as soil water content increases, and the release reaches a maximum when soil moisture is in the range of 45–55% maximum water holding capacity (WHC) (Bollmann and Conrad, 2004). N₂O emissions of denitrification in moist soils increase as soil water content increases, and maximum N₂O emissions were observed at 65–80% WHC (Bollmann and Conrad, 2004). Other studies found that denitrification is the major contributor of N₂O emissions under F–T environments, accounting for 80–90% of total N₂O emissions (Ludwig et al., 2004; Öquist et al., 2004; Yanai et al., 2007). Soil moisture in these studies was mostly maintained at 60% WFPS, similar to our W² group. In waterlogged soil, the N₂O fluxes maintained at a low level might be caused by an increased ratio of N₂/N₂O in the anaerobic environment (Teepe et al., 2004). Note that more N₂O is reduced to N₂ under high soil moisture. We also observed that the water level of the soil surface in W⁴⁵ increased about 0.5–1.0 mm compared with the initial value (Figure 7). Previous studies have indicated that unfrozen soil water would be transferred from deep soil to upper layers during freezing (Dai et al., 2019). Thus, the increased water layer enhances the physical barrier and impedes soil N₂O outflow. It is inferred that when the soil water was below the threshold value, the positive correlation between N₂O flux and moisture was due to the accumulation of N₂O, as the by-product of nitrification and the product of nitrate are reduced by denitrifying bacteria (Ji et al., 2015). However, complete denitrification accelerates N₂O consumption, and aqueous barriers alleviate N₂O emissions when the moisture is high.

The soil N₂O emissions with moderate moisture under low thawing temperature (T⁵) were much higher than in normal (T¹⁰) and high temperature (T²⁰) treatments. Surprisingly, the peak flux of W³⁰T⁵ was as high as 5,849.54 μg⋅m^–2⋅h^–1, which was the highest compared to all previous studies. Teepe et al. (2001) confirmed that frequent F–T cycles create abundant aerobic, anaerobic, or intermediate microzones, which are favorable for nitrification and denitrification. Consistent with Teepe et al. (2001), we observed that the frozen soil of W³⁰T¹⁰ and W³⁰T²⁰ samples were thawed completely after 12 h, but W³⁰T⁵ only thawed about 1/4 volume. Although we did not have direct evidence of the soil porosity changes under different thawing temperatures, it is not inconceivable that the air, free water, and ice crystals coexist in semithawed soil (Figure 6). We speculate that under moderate moisture and low thawing temperature, F–T cycles allowed for unique redox conditions favorable to N₂O production in nitrification and denitrification, and the semithawed state also provides large soil voids and ice openings for N₂O to move to the atmosphere.

In this study, the N₂O emissions of the W¹⁵ and W³⁰ groups, respectively, peaked at the 3rd/4th cycles and 13th cycles, similar to previous studies (1st–9th cycles) (Teepe et al., 2001; Bollmann and Conrad, 2004; Wu et al., 2019). Nutrient availability is widely regarded as the main reason for the massive release of N₂O during short F–T events (Risk et al., 2013; Congreves et al., 2018). This hypothesis holds that fragmented soil aggregates (Chai et al., 2014) and microbial mortality during F–T can release small molecular organic N (NH₄⁺, NO₃^–, and amino acid) (Yanai et al., 2011; Yu et al., 2011). The abundant metabolic substrates stimulate microbial activities and accelerate N₂O production. Our previous study also demonstrated that the available N explained 30% of the variation in N₂O fluxes during spring thawing (Chen et al., 2018). Subsequent studies also confirmed that dissolved organic matter and inorganic N jointly regulate N₂O flux from soils with different moistures during a freeze–thaw period (Wu et al., 2019).

Accordingly, it is easy to understand the sharp increase in emissions during the 1st–13th F–T cycles of W¹⁵ and W³⁰ treatments. However, previous studies showed that available N increased rapidly only in the first to fourth F–T cycles, and then, the contents stabilized after multiple repeated cycles (Grogan et al., 2004; Yu et al., 2011). This means that the promotion of N₂O emissions due to the increase in effective N over long-term F–T cycles is limited. However, the emission period of our study lasted until the 27th (W¹⁵) and 48th (W³⁰) cycles. Moreover, the cumulative emissions of W¹⁵ and W⁴⁵ were still increasing after 80 F–T cycles. Contrasting with the short-term effects of F–T on other seasonally frozen soil, frequent F–T disturbance may cause longer N transformation in alpine grasslands. Therefore, we suspected that the nutrient release in dry soil persisted during the long-term F–T cycles, and the N conversion pattern was different from that in soil with high moisture. The effect of F–T on the decomposition of soil organic matter in alpine grasslands under different water gradients needs further study.

This model (Figure 6) can provide some guidance for predicting the response of N₂O emissions in the nongrowing season under climate change, especially during spring thaw or snow melt. As our results showed, F–T disturbances drive N conversion and accelerate greenhouse gas (N₂O) emissions. Therefore, exploring the response of the soil C and N cycles to F–T events under different ecosystems of hydrothermal coupling is also necessary to predict the impacts and feedback of climate change in cold regions. Under the current climate, Wang et al. (2013) found that N₂O emissions of alpine meadow soil during spring thaw were 6.79 mg m^–2, accounting for 11% of emissions for the year. This in situ site is 30 km away from our sampling area. Although the value is much lower than our results, the warmer climate results in the topsoil undergoing many more F–T cycles and consequently prolongs the N₂O emission period in spring thawing and increases the proportion of emissions during the nongrowing season.

The current research shows that the climate of the QTP has become warmer and wetter, and the increased precipitation is mainly concentrated in the growing season (Zhang Y. H. et al., 2016). This means that if the rainfall/snowfall increased in late September or early March, as our results suggested, this change would probably aggravate soil N₂O emissions in dry ecosystems, such as meadow and steppe. The large gaseous N loss may further reduce N availability and even limit N supply during the wet season. However, the annual total evapotranspiration (522.28 mm) of this region was larger than precipitation (447.30 mm), and in winter, almost all snowfall was returned to the atmosphere directly via evapotranspiration. Moreover, the temperature increase in winter (0.45°C 10 a^–1) is nearly twice that in summer (Qin, 2014). Thus, topsoil would become drier in the nongrowing season. According to our prediction, dry soil would have reduced N₂O loss during F–T compared with wet soil. N₂O release from alpine grasslands may thus be a negative feedback to global warming. However, it cannot be ignored that the lower water levels of the alpine swamp in winter in the QTP possibly accelerated the succession of the swamp into a meadow or steppe ecosystem (Levy, 2008; Liu et al., 2018). This consequently exposed more peat soil and converted waterlogged soil to soil with a moderate moisture level.

It is estimated that the alpine meadow, steppe, and swamp in the QTP have approximately 133.52 Pg of organic C in the 0–0.75 m profile soils (Wang et al., 2002) and 16.08 Pg total N (Liu et al., 2012). Once the barrier of water and/or ice layer was lost in swamp soil, we detected that more than 1,000 times the normal level of N₂O would be released during F–T cycles. The large N pool size together with significant F–T cycles suggests a high risk of N₂O emissions and positive climate feedback across the alpine swamp. A past work has established that precipitation overrides warming in mediating soil N pools in an alpine grassland ecosystem on the QTP (Lin et al., 2016), but these studies are based on the growing season, and the mechanisms of F–T cycles on the soil N pools when the soil moisture and temperature change simultaneously in alpine ecosystems are still not clear.