We conducted parallel experiments in two contrasting temperate grassland ecosystems: a meadow steppe and a sandy steppe (Table 1; Fig. 1). Soil bulk density, texture, pH, total C and inorganic N were determined from our own field measurements and laboratory analysis of soil samples collected during the study establishment. The meadow steppe was situated at the Hulunber Grassland Ecosystem Observation and Research Station in northeastern China (49°19^′ N, 120°02^′ E, 628 m), while the sandy steppe was located at the Ordos Sandy Grassland Ecology Research Station in northern China (39°29^′ N, 110°11^′ E, 1290 m).

Both sites have a continental climate. The mean annual precipitation is 420 and 310 mm, and the mean annual temperature is -2 and 6.5 °C in the meadow steppe and the sandy steppe, respectively (http://data.cma.cn/, last access: 6 January 2021; https://www.ncei.noaa.gov/maps/hourly/, last access: 6 January 2021). The non-growing season for the meadow steppe extends from late September to late April of the following year, with a spring freeze-thaw period occurring from late March to late April. In contrast, the non-growing season for the sandy steppe lasts from mid-October to late March, with the spring freeze-thaw period occurring from late February to late March. During the study period, the meadow steppe had a persistent snow cover that reached a depth of 20–25 cm in late winter (January–February). In contrast, the sandy steppe exhibited shallower and more variable snowpack (typically 10 cm depth) due to higher wind redistribution and lower moisture retention. Under natural conditions, the meadow steppe in this study experienced a total of 19 freeze-thaw cycles, while the sandy steppe experienced 21 freeze-thaw cycles in early spring (http://nm.cma.gov.cn/, last access: 6 January 2021).

The meadow steppe features high plant diversity and fertile soils, while the sandy steppe exhibits lower diversity and nutrient-poor, coarse-textured soils (Table 1). This contrast enables a comprehensive assessment of FTC impacts across varying resource availability and community structures, as evidenced by significant baseline differences in N dynamics between sites. According to the Chinese Soil Classification (GB/T 17296-2009), the predominant soil type is loam in the meadow steppe and sandy loam in the sandy steppe. The meadow steppe soil has higher C and N content but slightly lower pH compared to the sandy steppe soil (Table 1). In the meadow steppe, the dominant plant species are Stipa baicalensis Roshev (perennial bunch grass), and subordinate species are Leymus chinensis (Trin.) Tzvel (perennial rhizome grass) and Carex pediformis C. A. Mey. (perennial forb), which together cover approximately 85 % of the site. In the sandy steppe, the dominant species are

Corethrodendron fruticosum (Pall.) B. H. Choi and H. Ohashi (semi-shrub), and subordinate species are Cleistogenes squarrosa (Trin.) Keng. (perennial bunch grass), and Klasea centauroides (L.) Cass. (perennial forb), covering about 70 % of the site (Table 1). The complementary strengths of these ecosystems enable robust predictions about grassland responses to change winter climate regimes. The detail information was described in Table S1.

In late October 2020, eighteen 3 m × 3 m plots were established at each site, with a 3 m buffer between neighboring plots. The experiment employed a randomized block design with three treatments and six replicates per site: (1) control (ambient FTC), (2) intensified low-frequency FTC (LFTC; +6 times; conduct 6 additional freeze-thaw cycles), and (3) intensified high-frequency FTC (HFTC; +12 times; conduct 12 additional freeze-thaw cycles). These treatments were designed to simulate projected increases in winter FTC frequency under climate change scenarios.

The treatment levels were based on historical climate data showing approximately 20 natural FTC typically occur during winter and early spring at both sites (Table 1; https://data.cma.cn/, last access: 6 January 2021). According to the definition of freeze-thaw cycling, a freeze-thaw cycle is defined as the process in which soil temperature (0–10 cm) rises above 0 °C and then subsequently drops below °C (Yanai et al., 2007). Therefore, the intensified FTC correspond to total increases in 30 % (+6 times; conduct 6 additional freeze-thaw cycles) and 60 % (+12 times; conduct 12 additional freeze-thaw cycles) in the frequency of FTC during winter and spring seasons, respectively.

Within each plot, we established a fixed 1 m × 1 m subplot for ¹⁵N tracing. Building upon established ¹⁵N tracing approaches (Ma et al., 2020), we applied ¹⁵NH₄¹⁵NO₃ solution prior to the onset of winter soil freezing. A solution containing 600 mg ¹⁵N L⁻¹ of ¹⁵NH₄¹⁵NO₃ was injected into 100 holes with a syringe guided by a grid frame (1 m × 1 m), with each hole receiving 2 mL of the labeled solution. The total application per subplot was 200 mL, which is equal to 120 mg ¹⁵N m⁻². The added ¹⁵N was kept within the natural fluctuation range of inorganic N in the soil, approximately 7 %–10 % of background soil inorganic N levels. We injected water into control treatments instead of the ¹⁵N tracer, and there were no significant differences in plant/microbial N concentrations when compared to the ¹⁵N treatments. This indicates that the ¹⁵N application did not disrupt natural N cycling processes (Ma et al., 2018).

Based on recent 5-year climatic records, our initial FTC treatments were scheduled approximately 15 d prior to the natural spring FTC period (late winter). For the freezing-thaw manipulation, a closed-top tent (3 m length × 3 m width × 2 m height) was installed in each plot during each warming manipulation. The heating tents were constructed with polyester fabric, featuring sealed tops and mesh-sided windows to prevent excessive CO₂ accumulation while maintaining temperature control. Within each tent, we used a propane air heater (Mr. Heater, USA) to raise soil temperature to 2–3 °C (0–15 cm), maintaining this temperature continuously for 8 to 10 h each time. Continuous temperature logging was performed using a temperature detector per treatment positioned at 10 cm soil depth, with data recorded at half-hour intervals throughout the experiment. The temperature was then allowed to drop to approximately -2 °C over a period of 4 h to complete one freeze-thaw cycle, the 5 cm depth was periodically verified with a handheld thermometer specifically during FTC treatments to ensure target temperature thresholds were met. Two intensified FTC regimes were implemented: (i) high-frequency FTC (HFTC) with 12 additional cycles administered every 1–6 d, and (ii) low-frequency FTC (LFTC) with 6 additional cycles every 3–8 d. During the natural freeze-thaw period, all artificial FTC treatments were deliberately conducted when daily mean temperatures remained below -2 °C to avoid interference with natural cycles.

Statistical significance of treatment effects was assessed by one-way ANOVA. Differences between treatments were reported as non-significant at p>0.05 or significant at p<0.05. Repeated measures ANOVA was used to analyze the influences of different FTC treatments, sampling times, and grassland types on the measured indicators. Spearman correlation analyses were used as initial screening tool to identify relationships between environmental factors and plant ¹⁵N acquisition across different treatments. Random Forest analysis was then employed as a more robust machine learning method that can handle high-dimensional data and minimize overfitting, while effectively ranking variable importance and handling collinearity among predictors. Spearman correlation coefficients between variables were calculated using the rcorr function (in the Hmisc R package). To assess the relative importance of predictors for plant ¹⁵N acquisition capacity, a random forest model was constructed using the randomForest and rfPermute packages in R. The dataset was randomly partitioned into a training set (70 %) for model development and a testing set (30 %) for model validation. All above-mentioned analyses were conducted with SPSS 21.0 (IBM, Chicago, IL, USA) and RStudio 2025.5.0 (Posit Software, Boston, MA, USA). All graphics were generated using SigmaPlot 14.0 (Systat Software, Inc., San Jose, CA, USA), Origin 2021 (OriginLab Corp., Northampton, MA, USA) and RStudio 2025.5.0.

The edaphic conditions, including soil total C content, inorganic N content, and texture, exhibited significant differences between the two temperate grasslands (Table 1). Throughout the winter freezing period, the minimum soil temperatures (0–10 cm) were about -23 °C in the meadow steppe and -20 °C in the sandy steppe, respectively (Fig. 2a, b). In early spring, soil temperatures rose rapidly, accompanied by significant snowmelt. However, neither intensified LFTC nor HFTC had any significant impact on soil temperature in the subsequent growing season. In contrast, intensified low-frequency FTC (LFTC) and high-frequency FTC (HFTC) enhanced soil moisture by 0.03 and 0.05 m³ m⁻³, respectively, over much of the seasons (Fig. 2c, d).

Intensified HFTC significantly increased soil NH4+-N concentrations and net ammonification rates in both grasslands during spring, with the most pronounced effects observed in the meadow steppe (Fig. 3a, b, e, f). In contrast, soil NO3--N concentrations and net nitrification rates remained stable across all treatments (Fig. 3c, d, g, h).

Intensified HFTC significantly decreased the soil microbial biomass C (MBC) in spring, while the effect of HFTC on microbial biomass N (MBN) persisted to summer (Fig. 4a–d). In the meadow steppe, HFTC significantly decreased MBC by 16.2 % (Fig. 4a), while LFTC and HFTC significantly increased MBN by 26.2 % and 26.9 %, respectively (Fig. 4c). In the sandy steppe, HFTC significantly decreased MBC by 11.3 % in early spring. Unlike MBC, both LFTC and HFTC significantly increased MBN by 8.5 % and 28.2 %, respectively (Fig. 4b, d).

In contrast to the significant effects of HFTC, intensified LFTC had no significant impact on the shoot or root biomass N of the selected plant species at either site (Fig. 5a–f). In the meadow steppe, HFTC significantly increased shoot and root biomass N of Stipa baicalensis (perennial bunch grass) by 19.7 % and 21.8 % at the end of the growing season, respectively (Fig. 5a). In contrast, HFTC significantly reduced biomass N in the perennial rhizome grass Leymus chinensis (shoot: 23.9 %; root: 16.2 %) and the perennial forb Carex pediformis (shoot: 22.2 %; root: 18.0 %) (Fig. 5c, e). A similar differential response was observed in the sandy steppe. HFTC significantly increased biomass N in the semi-shrub Corethrodendron fruticosum (shoot: 22.6 %; root: 23.7 %) but significantly reduced it in the perennial bunch grass Cleistogenes squarrosa (shoot: -25.3 %; root: -12.1 %) and the perennial forb Klasea centauroides (shoot: -23.1 %; root: -20.3 %) (Fig. 5b, d, f).

In both grassland types, soil ¹⁵N recovery peaked in early spring, followed by a rapid decline from late spring to late summer. This was then followed by a gradual increase in recovery until late autumn (Fig. 6c, d). In contrast, plant ¹⁵N acquisition increased steadily throughout the growing season in both grasslands, while microbial ¹⁵N recovery exhibited only modest fluctuations over the entire growing season (Fig. 6e–h).

During the early growing season, intensified LFTC had no significant effect on total ¹⁵N recovery in soil-microbe-plant systems, while intensified HFTC significantly increased total ¹⁵N recovery (Fig. 6a, b). LFTC did not significantly impact soil ¹⁵N recovery, but HFTC significantly increased soil ¹⁵N recovery in the two grasslands Fig. 6c, d). In the meadow steppe, intensified LFTC and HFTC significantly enhanced microbial ¹⁵N recovery by 38.0 % and 26.6 %, respectively, and by 49.5 % and 32.5 % in the sandy steppe (Fig. 6e, f). In contrast to the positive effects on microbial recovery, HFTC significantly reduced plant ¹⁵N acquisition in both grasslands. LFTC had no significant effect on plant ¹⁵N acquisition (Fig. 6g, h). The leaching ¹⁵N (deepsoil: 30–50 cm) in meadow steppe was more than that in the sandy steppe. In both grasslands, neither LFTC nor HFTC had a significant effect on the leaching ¹⁵N (Fig. 6i, j).

In the meadow steppe, the ¹⁵N acquisition in the shoots of S. baicalensis (perennial bunch grass) and C. pediformis (perennial forb) were comparable, while L. chinensis (perennial rhizome grass) exhibited lower ¹⁵N acquisition. In contrast, the highest ¹⁵N acquisition in roots was observed in L. chinensis, followed by C. pediformis and S. baicalensis (Fig. 7a, c, e). In the sandy steppe, both shoot and root ¹⁵N acquisition of C. fruticosum (semi-shrub)  were the highest among the studied species. This was followed by the shoot ¹⁵N acquisition of C. squarrosa (perennial bunch grass) and K. centauroides (perennial forb). Notably, the root ¹⁵N acquisition of K. centauroides was higher than that of C. squarrosa (Fig. 7b, d, f).

HFTC significantly altered these acquisition patterns in a species-specific manner (Fig. 7). In the meadow steppe, HFTC significantly increased shoot and root ¹⁵N acquisition of S. baicalensis by 5.8 % and 9.3 %, respectively, but significantly decreased it in L. chinensis (shoot: 16.4 %; root: 12.1 %) and C. pediformis (shoot: 4.9 %; root: 7.8 %) (Fig. 7a, c, e). Similarly, in the sandy steppe, HFTC significantly increased ¹⁵N acquisition in C. fruticosum (shoot: 3.8 %; root: 18.4 %) but significantly reduced it in C. squarrosa (shoot: 16.7 %; root: 14.4 %) and K. centauroides (shoot: 16.1 %; root: 14.1 %) (Fig. 7b, d, f).

The correlation analysis revealed distinct and treatment-specific shifts in the relationships between plant ¹⁵N acquisition and environmental predictors across two grasslands (Fig. 8). In both grasslands, plant ¹⁵N acquisition exhibited the strongest positive correlation with microbial ¹⁵N retention under control treatment (Fig. 8a, b). In the meadow steppe, under both LFTC and control, microbial ¹⁵N retention, soil temperature, soil moisture and soil NO3--N levels were positively correlated with plant ¹⁵N acquisition (Fig. 8a, c). Under HFTC, plant ¹⁵N acquisition also exhibited a positive correlation with MBC (Fig. 8e). In the sandy steppe, under LFTC and HFTC, plant ¹⁵N acquisition exhibited the strongest positive correlation with microbial biomass N, followed by soil temperature, and soil NO3--N levels (Fig. 8d, f). Conversely, soil ¹⁵N retention, net nitrification rate, and net ammonification rate were negatively correlated (Fig. 8d, f). Under HFTC, plant ¹⁵N acquisition also exhibited a positive correlation with soil moisture (Fig. 8d).

Random forest analysis revealed that soil temperature and soil ¹⁵N retention were the primary predictors of plant ¹⁵N acquisition across all treatments (Fig. 9). Notably, MBC and NH4+-N levels did not emerge as a significant predictor in any of the models. However, the importance of other predictors varied between grasslands and treatments. In the meadow steppe, the control, LFTC, and HFTC treatments each retained five key predictors. Dominant predictors under LFTC included net ammonification rate, soil NO3--N levels and microbial ¹⁵N retention. Under HFTC, key predictors shifted to soil moisture, microbial ¹⁵N retention and net ammonification rate. Neither net nitrification rate nor MBN emerged as significant predictors under LFTC or HFTC (Fig. 9c, e). In the sandy steppe, net nitrification rate and soil moisture were key predictors under both LFTC and HFTC (Fig. 9d, f). Both LFTC and HFTC exhibited more predictors than control. Under LFTC, MBN was not a significant predictor (p=0.089), under HFTC, net ammonification rate and microbial ¹⁵N retention were also not significant predictors.

Our study reveals that intensified FTC triggers a strategic shift in soil microbial nutrient use, characterized by a notable decoupling between microbial C and N dynamics (Fig. 4). The significant reduction in MBC during the early growing season aligns with the previous observations of microbial lysis and physiological stress induced by freeze-thaw events (DeLuca et al., 1992; Walker et al., 2006). A critical finding was the stability or even increase in MBN under C-limited conditions in early spring (Fig. 4), indicating a decoupled microbial response. We propose this reflects a microbial adaptation to prioritize N retention. Faced with an inorganic N pulse from cell lysis and aggregate disruption (Fig. 3) yet constrained by C scarcity, microbes engage in luxury N immobilization. This strategy allows them to secure and store N, preventing its loss from the system (Christopher et al., 2008; Skogland et al., 1988; Wang et al., 2024). This physiological trade-off maintains ecosystem N retention at the expense of C use efficiency (Schimel and Bennett, 2004; Yu et al., 2011). Therefore, the microbial response to FTC is one of strategic re-allocation, shifting their stoichiometry to optimize N storage during a critical window of availability and instability.

Contrary to our first hypothesis, intensified FTC did not increase lead to ecosystem-level losses of the total ¹⁵N tracer in either temperate grasslands. Instead, high-frequency FTC (HFTC) significantly enhanced total ¹⁵N recovery within the soil-microbe-plant system during the early growing season (Fig. 6a, b), indicating that effective conservation mechanisms were activated. This finding challenging the prevailing paradigm that winter climate change inevitably promotes widespread N loss (Han et al., 2018; Song et al., 2017).

This observed retention capacity can be explained through three interconnected mechanisms. First, the soil pool acted as a major and persistent sink. The significantly elevated soil ¹⁵N retention under HFTC (Fig. 6) points to the efficient physical protection and chemical stabilization of the released N. This protection likely occurred through incorporation within stable soil aggregates and adsorption onto organic matter surfaces, reducing N mobility and availability for loss pathways (Bhattacharyya et al., 2019).

Second, soil microbes served as a crucial biological buffer during the critical early spring period. The significant increase in microbial ¹⁵N immobilization during early spring (Fig. 6) indicates their rapid capture of winter-derived N. Crucially, this microbial immobilization occurred when plant uptake was minimal, thereby securing the N pulse during this vulnerable window (Turner and Henry, 2009; Zheng et al., 2024). Third, FTC-induced increases in soil moisture mediated ¹⁵N availability. Our random forest analyses identified soil moisture as a significant predictor of plant ¹⁵N acquisition (Fig. 8). The elevated moisture under FTC treatments (Fig. 2b) likely enhanced N mobility, facilitating diffusion to roots. This moisture-driven promotion of N flux created favorable conditions for plant uptake, yet within the framework of effective ecosystem retention as evidenced by the absence of significant leaching losses.

While the stability of the microbial ¹⁵N pool over time indicates limited direct transfer of immobilized N to plants, its role in initial N stabilization during the vulnerable early spring period was paramount. Subsequent plant ¹⁵N uptake likely derived from other soil pools replenished by mineralization, indicating a decoupling of the typical synchrony between microbial and plant N partitioning.

Contrary to our first hypothesis, total ¹⁵N recovery was statistically similar between the two contrasting grassland ecosystems under intensified FTC conditions (Fig. 6). This convergence in ecosystem-level ¹⁵N retention can be explained by several compensatory mechanisms: First, while the meadow steppe exhibited higher net N mineralization rates in early spring, releasing a larger initial nitrogen pulse, the sandy steppe compensated through more efficient microbial and plant uptake of the N sources. This was evidenced by significantly lower soil NH4+-N concentrations in the sandy steppe (Fig. 4), suggesting adaptation for rapid N acquisition in this resource-limited system. Consequently, both ecosystems achieved statistically similar ¹⁵N levels in microbial and plant pools despite their divergent soil conditions (Fig. 4e–h; Table 1).

Second, hydrological pathways of winter-derived N loss were similarly constrained in both grasslands. The minimal ¹⁵N levels detected in deep soil layers (30–50 cm) (Fig. 5e, f) indicate limited leaching losses, demonstrating that intensified FTC did not disproportionately enhance N loss in the coarser-textured sandy steppe. This established a similar baseline of physical N conservation in both systems. Therefore, the similar levels of ecosystem ¹⁵N retention were not achieved through identical processes, but through different yet effective strategies in plant N uptake, physical conservation, and microbial immobilization.

Our results strongly support the second hypothesis that intensified FTC alter species-specific acquisition of winter N sources. While HFTC significantly reduced ¹⁵N acquisition at the community-level, this overall trend concealed strongly species-level divergence (Fig. 7). This divergence was not random but was clearly aligned with key plant functional traits, particularly spring phenology and root system architecture (Table S1).

The enhanced ¹⁵N acquisition under HFTC by dominant species, S. baicalensis in the meadow steppe and H. mongolicum in the sandy steppe, exemplifies a trait-based strategy for exploiting freeze-thaw induced N pulses (Table S1). In the meadow steppe, S. baicalensis capitalized on its early spring growth and dense root morphology (Ma et al., 2018) to dominate N acquisition. The high root density provided a superior absorptive surface area in the topsoil, where FTC-mobilized N was concentrated, granting it a competitive advantage over species with coarser or less-developed root systems. In the sandy steppe, the deep-rooted legume C. fruticosum (Lonati et al., 2015) buffered against surface perturbations by accessing stable subsurface N and water. The success of these species underscores that the coupling of early phenology or deep resource access with robust root systems is a critical adaptation to FTC-induced stress, allowing them to effectively monopolize winter N resources (Miller et al., 2009).

In contrast, subordinate species (L. chinensis, C. pediformis, C. squarrosa, K. centauroides) showed significantly decreased ¹⁵N acquisition, a consequence of their phenological and architectural mismatch with the FTC-altered regime. Their later phenology likely prevented utilization of the early N pulse, while shallow, damage-susceptible root systems further constrained access to winter N sources (Table S1; Ma et al., 2018). This competitive disadvantage arose through two interconnected mechanisms. First, phenological asynchrony placed the subordinate species at a critical disadvantage. The early-season N pulse released by HFTC occurred before these later-active species had initiated substantial root activity or shoot growth (Table S1). Consequently, they missed the peak window of N pulse, which was preemptively captured by early-season competitors.

Second, structural vulnerability exacerbated their disadvantage. The fine, shallow root systems of perennial forbs, particularly C. pediformis and K. centauroides, are highly susceptible to HFTC-induced root damage (Table S1; Campbell et al., 2014; Ye et al., 2017). This vulnerability was supported by our data showing the significant reduction in root biomass N for these species (Fig. 5). Such damage not only increased fine root mortality but also further constrained their capacity to access winter-derived N (Hosokawa et al., 2017; Reinmann et al., 2019).

Ecologically, the divergent responses among plant species can be primarily attributed to a disruption of temporal niche partitioning. HFTC generate an early-season N pulse that preferentially favors species with pre-existing adaptations to cold-season conditions, such as early spring phenology and robust root systems. This initial advantage is further amplified by the greater susceptibility of later-active species to root damage, thereby intensifying competitive asymmetry and potentially driving long-term shifts in plant community structure. Despite the stability of soil and microbial N pools, the overall reduction in community-level ¹⁵N acquisition under HFTC suggests a potential decoupling between ecosystem N retention and plant N utilization. This indicates that ecosystem resilience, defined as the capacity to maintain both structure and function, may be compromised, as the ability to conserve N does not necessarily ensure unchanged patterns of plant resource acquisition.

This study provides valuable insights into ecosystem N cycling under intensified FTC, yet several limitations should be acknowledged. First, while our ¹⁵N tracer approach precisely tracked the fate of winter-derived inorganic N, it did not capture dynamics of the native soil N pool, particularly mobilization and loss pathways of unlabeled organic N. Second, the temporal resolution of our sampling, while appropriate for quantifying seasonal patterns of plant N uptake, was insufficient to capture rapid microbial N transformations and gaseous fluxes occurring within days following FTC events. Third, while sampling the 0–20 cm soil layer captured the majority (70 %–80 %) of the root systems, it may not fully represent the absolute ¹⁵N acquisition by deep-rooted species was likely underestimated. Finally, due to equipment constraints, we did not monitor photosynthetically active radiation (PAR) or precise CO₂ levels within experimental tents; including these parameters in future studies would offer a more comprehensive understanding of microclimatic perturbations.

Building on these limitations, we propose two key priorities for future research: First, pinpoint the sources of newly available N during FTC. It would be valuable to differentiate the specific origins of newly available N during FTC, whether derived from microbial cell lysis, root mortality, or physical disruption of soil aggregates. Clarifying these sources is essential to accurately trace the pathways and retention mechanisms of FTC-mobilized N. Second, conduct high-frequency monitoring of greenhouse gas fluxes. Simultaneous monitoring of greenhouse gases (particularly N₂O and CO₂) with high temporal resolution during FTC events is crucial. This approach would help elucidate the coupling of microbial C and N cycling, especially given that FTC-induced N₂O peaks often occur without corresponding CO₂ increases, a phenomenon potentially related to the decoupled responses of microbial biomass C and N observed in our study. By systematically addressing these research priorities, we can significantly advance the mechanistic understanding of N cycling and ecosystem responses to winter climate change.