Field sampling and observation experiments were conducted from April 2021 to July 2022. The study site (Figure 1) was located in the Shuanglong Forest Farm (35°39′ ~ 35°43’N, 108°56′ ~ 108°58′E), a natural secondary forest in the Ziwuling Mountains of North China (Zhang et al., 2022). Our study site was 100 × 100 m. The climate of this site was a warm temperate semihumid climate, with a mean annual temperature of approximately 7.4°C and a mean annual precipitation of 587.6 mm (Chai et al., 2016). The main soil type was loessial soil, which was turbid brown or orange. The soil texture was loose and soft with few roots and pores, which indicated silt loam. The typical arbor species include Betula platyphylla, Swida macrophylla, Carpinus turczaninowii Hance, Quercus aliena Bl, Quercus liaotungensis, Rhus potaninii Maxim, and cer davidii Franch. The typical shrub species include Acer tataricum subsp.ginnala, Viburnum dilatatum Thunb, Cotoneaster multiflorus Bge, Rhamnus leptophylla Schneid, Lonicera hispida Pall. ex Roem. et Schult.

Three sites with flat terrain were selected as sampling points for the measurement of the R_S rate (μ mol m⁻² s⁻¹). We installed ACE automatic R_S monitoring systems on iron rings with the inner diameter of 20 cm and the height of 10 cm (ACE-200, Ecotech Ecological Technology Ltd) and inserted 4–5 cm into the soil at each sampling point. R_S measurement sites were chosen to be more than 50 cm away from the surrounding vegetation, with each site being more than three meters away. To reduce soil disturbance, we inserted the iron rings at least 24 h before the measurement, and the broken roots and litter on the soil surface were removed. From April 2021 to July 2022, we used automatic R_S monitoring systems to monitor the R_S rate every 30 min for all sample points every month for 24 h. The R_S monitoring systems could simultaneously measure and record the soil temperature (T_S, °C) and soil moisture (M_S, %vol) within 0–10 cm below the surface soil of the sampling site. A meteorological monitoring station (CR200Series) was established at the research site to collect air temperature and moisture data from April 2021 to July 2022.

During the study period, from April 2021 to July 2022, soil sampling was performed every 2 months. Five sampling points were set at the research site using the five-point sampling method. To avoid edge effects, the sampling points were neither close to the edge of the plot nor far from the edge of the sample plot. Each sampling site was 5 × 0.5 m plot. The sampling sites were surrounded by abundant vegetation and the soil surface had obvious humus layers. We divided each sampling point into 12 areas, assigned a random block to all the sampling points, and conducted sampling according to the random block order (Supplementary Figure S1). The sampling depth was divided into two types. The soil at a depth of 0–30 cm below the surface soil was used as topsoil, and the soil at a depth of 30–100 cm was used as subsoil. A total of five replicates were collected separately for topsoil and subsoil. The soil from each depth at each sampling point was mixed evenly after collection, and the broken roots and litter in the soil were removed to reduce errors in the analysis process. Sterile gloves were worn during soil collection to prevent soil contamination. The soil samples were transported in sterile sampling bags, stored in a freezer, and taken to a laboratory by car for further analysis.

We used the Kjeldahl method (Bradstreet, 1965) to determine the total soil nitrogen (TN), and the soil was hydrolyzed under alkaline conditions in a diffusion dish (Wang, 2010) to calculate the content of alkaline hydrolyzed nitrogen (HN). We used the alkali fusion-Mo-Sb antispectrophotometric method (Chen et al., 2018) to determine the total phosphorus (TP) and sodium bicarbonate solution (Cade-Menun and Lavkulich, 1997) to determine the available phosphorus (AP). The soil-available potassium (AK) was determined by ammonium acetate flame photometry (Zanati et al., 1973).The potassium dichromate oxidation-external heating method was used to determine soil organic matter (SOM), and then the SOM was determined by titration with a standard ferrous iron solution (Zhu et al., 2020). The soil pH was measured by using a pH meter. Microbial biomass carbon (MBC), microbial biomass nitrogen (MBN), and microbial biomass phosphorus (MBP) were determined by using the chloroform fumigation extraction method (Vance and Brookes, 1987). The soil samples were leached with KCL solution and then analyzed using a continuous flow analyzer to determine NH₄⁺--N and NO₃⁻--N (Liu et al., 2014).

We used a fluorometric method (Eivazi and Tabatabai, 1988) to measure the β-1,4-glucosidase (βG) activity in the soil and a nitrophenol colorimetric method (Wood and Bhat, 1988) to measure the cellobiohydrolase (CBH) activity. Polyphenol oxidase (PPO) was determined spectrophotometrically by using pyrogallol (1,2,3-trihydroxy benzene) as a substrate (Bach et al., 2013). Peroxidase (PER) was measured by calculating the rate of substrate oxidation after the addition of H₂O₂ (Burns et al., 2013).

The R_S mean value and error were calculated from three replicate measurements. The mean values and errors of soil physicochemical and microbial properties were calculated from five replicate measurements. Pearson correlation analysis was used to examine the correlation of R_S with T_S and M_S. Origin 2017 software was used to obtain the regression equations between R_S rate, T_S, and soil M_S, and then these regression relationships were plotted. Monthly and seasonal differences in R_S and soil physicochemical and microbial properties were tested by ANOVA. The datasets were checked for normality and homogeneity assumptions before performing ANOVA. The magnitude of this feedback largely depends on the temperature sensitivity of SOM decomposition (Q₁₀).

Q₁₀ was measured by the exponential relationship between R_S and T_S and was calculated as follows:

where T_S is the soil temperature, a is the R_S rate when the soil temperature is 0°C, and b is the temperature coefficient reflecting the temperature sensitivity of R_S.

To examine how soil microbial characteristics influenced R_S, structural equation modeling (SEM) was performed with Amos software (IBM SPSS Amos 26.0.0) for different soil layers. In stepwise multiple regression (Supplementary Tables S1, S2), in order to optimize the model, we removed the non-significant variables and paths. We evaluated the goodness of fit of the model according to the low chi-square (χ²; the model is a great fit when 0 ≤ χ²/df ≤ 2) (Tabachnick and Fidell, 2007), the high whole-model p value (if p > 0.05, there is no path loss and the model was a great fit), the comparative fit index (CFI; the model is a great fit when 0.97 ≤ CFI ≤ 1) (Hu and Bentler, 1999), and a root mean square error of approximation (RMSEA; the model is a great fit when 0 ≤ RMSEA ≤0.05) (Vile et al., 2006).

The monthly variations in air temperature and air moisture are shown in Figure 2. During the observation period from April 2021 to July 2022, the average air temperature was 16.1°C, the highest temperature was 22.2°C, and the lowest temperature was 4.9°C. The average air moisture was 61.7%, the maximum moisture was 75.68%, and the minimum moisture was 46.70% (Figure 2A).

During the observation period, the T_S showed a pattern consistent with the seasonal variations in air temperature and air moisture. That is, the T_S gradually increased from April to July 2021, reaching a maximum value of 19.4°C in July 2021, and then the T_S decreased for the rest of the year. The M_S had obvious monthly variations during the observation period, reaching a maximum value in October 2021 and minimum value in June 2021 (Figure 2B). The maximum and minimum values were 92.44 and 42.20%, respectively.

The soil at the study site was alkaline, and there was no significant difference in pH between the topsoil (0–30 cm) and the subsoil (30–100 cm) (Table 1). The TP content in the subsoil was significantly higher than that in the topsoil, while the other soil physicochemical properties in the subsoil were lower than those in the topsoil, and AK and SOM were significantly reduced in the subsoil (p < 0.01).

The diurnal variation in the R_S rate in different months is shown in Supplementary Figure S2. The R_S rate showed a multi-peak distribution trend, and the R_S rate reached its peak at noon every day except in April and May within the 24 h observation period of each month. Except in June and August 2021, the R_S rate showed a minimum value in the morning (approximately 5:00 to 8:00 AM), and then the R_S rate gradually increased. The monthly variation in the R_S rate showed a trend of increasing and then decreasing. The R_S rate gradually increased after April, reached a maximum in July 2021, and then gradually decreased (Supplementary Figure S3).

The seasonal changes in the R_S rate are shown in Figure 3. On the seasonal scale, there were significant differences between the R_S rates in different seasons. In 2021, the average R_S rate in summer was 0.75 μmol m⁻² s⁻¹, which was significantly higher than that in spring (0.52 μmol m⁻² s⁻¹) and autumn (0.37 μmol m⁻² s⁻¹) (p < 0.05). In 2022, the average R_S rate in summer was 0.97 μmol m⁻² s⁻¹ (Figure 3).

The correlation analysis results showed that there were significant correlations between T_S, M_S, and R_S (p < 0.01). The relationship between T_S and the linear equation fitting the diurnal-scale variation in R_S is shown in Figure 4, and the relationship between M_S and the linear equation fitting the diurnal-scale variation in R_S is shown in Supplementary Figure S4. The R_S rate increased with T_S and decreased with M_S. According to Eq. (1), Q₁₀ is 2.61, which is within the normal range (Zhou et al., 2009).

In the topsoil, the NO₃⁻--N from August 2021 to July 2022 was significantly higher than that from April to June 2021, and the MBC from October 2021 to July 2022 was significantly higher than that from April to August 2021. NH₄⁺--N reached a maximum (12.59 mg/kg) in August 2021, and there were no significant differences in NH₄⁺--N between other months except in August 2021. The MBP in the topsoil in August 2021 was significantly higher than that in the other months. The MBN reached the maximum (34.95 mg/kg) in May 2022 (Figure 5A). In the subsoil, the NO₃⁻--N from April 2021 to June 2021 was significantly higher than that in other months, and there was no significant difference in NH₄⁺--N among the 6 months. The MBC reached a maximum (113.88 mg/kg) in October 2021. The MBN from May to July 2022 was significantly higher than that from April 2021 to August 2021. There were no significant changes in MBP from April 2021 to July 2022 (Figure 5B).

The soil enzyme activity in the topsoil and subsoil varied significantly among different months (p < 0.05). The PER activity in the topsoil in May 2022 and July 2022 was significantly lower than that in the other months, and in the subsoil there were significant differences in the PER activity between April 2021 and July 2022. The PPO activity in both the topsoil and subsoil varied significantly among different months. Similar to that in the topsoil, in the subsoil, the βG activity in May 2022 was significantly higher than that in other months. The βG activity in the topsoil in April, June, and August 2021 was significantly different from the βG activity in the subsoil. The CBH activity in the topsoil in August 2021 was significantly lower than that in other months, and in the subsoil, the CBH activity in May 2022 was significantly higher than that from April to August 2021 but not significantly different from that in other months. In addition, from April 2021 to August 2021, the CBH and βG activities were significantly different in the topsoil and subsoil, so we believe that the CBH and βG activities in the topsoil were generally greater than those in the subsoil (Table 2).

Based on the stepwise multiple regression results (Supplementary Tables S1, S2), we determined the variables that mostly explained the variation in R_S. Model optimization was performed continually until the model fits well. SEM demonstrated the influence of R_S in different soil layers (Figure 6). The model for topsoil showed values of χ² = 2.809, p = 0.422, df = 3, RMSEA = 0, and CFI = 1 (Figure 6A); the model for whole soil showed values of χ² = 4.339, p = 0.362, df = 4, RMSEA = 0, and CFI = 1 (Figure 6B).

In the topsoil, T_S, βG, MBP, and PPO all directly affected R_S, except for MBP, T_S, βG and PPO, which were significantly and positively correlated with R_S, and T_S, βG were the variables that had the strongest effects on R_S. In the whole soil, T_S, βG, MBP, MBN, and PPO all directly affected R_S, except for MBP, T_S, βG, PPO, MBN, which were significantly and positively correlated with R_S, and T_S, βG, PPO were the variables that had the strongest effects on R_S.

Global warming not only increases the temperature of the atmosphere but also leads to changes in precipitation, which in turn causes greater variation in T_S and M_S (Zhang et al., 2016). In this study, the air temperature reached its highest value in summer and then gradually decreased, and the atmospheric moisture showed a multipeak trend, which experienced a decrease followed by an increase in the summer (Figure 1A). This may be related to the specific climate of the Loess Plateau. Since the Loess Plateau is an area sensitive to climate change, changes in atmospheric temperature and precipitation caused by global warming often lead to frequent droughts in many areas (Piao et al., 2010a). The climate of the Loess Plateau showed a trend of aridity in spring and summer (Hou et al., 2021) and then experienced violent precipitation in autumn, causing the air temperature to decrease after August, while air moisture began to increase significantly after August, reaching a maximum in October, and then gradually decreased again. T_S and M_S also show similar trends to air temperature and moisture (Figure 1B).

It has been demonstrated in previous studies that R_S is closely related to T_S (Wang et al., 2006). During the observation period, the monthly diurnal variation in R_S showed a multi-peak trend, and the R_S rate reached its peak value at noon and decreased to a minimum value in the morning (Supplementary Figure S2). This is consistent with the results of Wang et al. (2018). A possible explanation for this phenomenon is that the reaction process of R_S is mainly catalyzed by soil enzymes, and temperature is the main limiting factor affecting soil enzyme activity (Melillo et al., 2018). When the T_S is low, the activities of some critical enzymes that control R_S decrease, resulting in a lower R_S rate. On the monthly scale, with increasing T_S, the R_S rate starts to increase in April, reaches a maximum value in July, and then gradually decreases to a lower value in October, which is consistent with the results of Tong et al. (2021). Our results differ from the results of Wen et al. (2018). A possible explanation for this phenomenon is the sensitivity of the research site to climate warming, resulting in a significantly higher rate of temperature change than in the other areas (Cao et al., 2016), thereby increasing the T_S to a maximum at an earlier time and leading to an increase in the R_S rate. The temperature sensitivity (Q₁₀) of R_S is often used as an important parameter to measure the feedback of R_S to global warming (Hu et al., 2013). During the whole study period, there was a significant statistical relationship between the R_S rate and T_S (p < 0.01; Figure 4), and the fitting effect between the R_S rate and T_S also had some explanatory significance (R² = 0.2654). Increased temperature stimulates R_S because climate warming may enhance the activity of soil microorganisms and promote soil organic C and litter decomposition, which partly explains why the R_S rate in summer is significantly higher than that in spring and autumn (Figure 3). Our findings suggest that a sustained increase in temperature may lead to greater soil C loss; that is, climate warming reduces soil C sinks.

M_S has been identified in previous studies as a major factor affecting R_S, especially after drought rewetting events stimulate R_S (Hu et al., 2019). Our results show that M_S is significantly negatively correlated with R_S (Supplementary Figure S4), proving that M_S limits CO₂ emissions from soil in a shorter period, which is different from the results of Yu et al. (2021). There may be several reasons for this: first, higher M_S inhibits the CO₂ transportation process from the atmosphere to the soil (Yan et al., 2018). During the study period, the M_S variation range was 92.44% ~42.20%, especially in summer and autumn, and the M_S remained at a high level (Figure 2B). It has been suggested in previous studies that under anoxic conditions, soil organic carbon (SOC) may be more persistent (Li et al., 2021), resulting in a decrease in R_S rate. Second, differences in SOC and microbial communities that decompose specific soil SOC lead to different relationships between M_S and R_S (Yan et al., 2018). Third, in this study, M_S may not be the main factor affecting the R_S rate because R_S is often regulated by multiple factors (Duan et al., 2021). For example, soil salinity and R_S show a negative correlation. When in a salt-stress environment, the activities of plant roots and soil microorganisms are severely affected (Song et al., 2021).

The correlations between T_S, M_S, and R_S demonstrate that T_S and M_S data are useful for optimizing microbial decomposition models, which facilitates soil microbial activity prediction much better in the context of future climate change.

Most previous studies have focused on the topsoil physicochemical properties (Liu et al., 2021). In this study, there were significant differences in the TP, AK, and SOM contents of different soil layers; TP was higher in the subsoil, and AK and SOM were higher in the topsoil (Table 1). The soil physicochemical properties changed with increasing soil depth, and our results are consistent with the results of Rahman et al. (2022). Microbial secretions can significantly affect soil potassium availability (Zorb et al., 2014), and factors such as microbial abundance and activity determine the pathway of soil litter conversion into SOM (Witzgall et al., 2021). In addition, plant roots and soil microbial communities can also dissolve P in the soil (Yang et al., 2021). Litter and most of the plant roots in the study site are concentrated in the topsoil. As shown in Figure 7, the microbial enzyme activity in the topsoil is higher than that in the subsoil, which proves that the microorganisms in the surface soil may be more active. Therefore, the AK and SOM contents were significantly higher than those in the subsoil, and the TP content was significantly lower than that in the subsoil. C and N cycle and nutrient turnover in soil are carried out by microorganisms through substrate (organic matter and litter) decomposition. In recent years, microbial decomposition models have been commonly used to explore the role of soil microorganisms in the coupled C and N cycle (Wang et al., 2014; Buchkowski et al., 2015). Our results can provide a reference for describing C, N, and P stocks and stoichiometry as well as soil nutrient distribution patterns in Loess Plateau soils and provide initial response data for soil microbial decomposition models.

The soil microbial biomass can regulate microbial biochemical processes and nutrient cycling and affect soil physical and chemical properties, which in turn affects soil quality (He et al., 2003). MBC is often considered to be a crucial factor affecting R_S, reflecting the ability of microorganisms to utilize SOC (Zeng et al., 2018). We found that, whether in the subsoil or topsoil, the MBC content showed a gradually increasing trend, and the MBC in May 2022 was significantly higher than in the other 3 months (Figure 5). We found that the trend of R_S was out of sync with that of the MBC. R_S decreased to a minimum value in autumn 2021, but the MBC reached a peak value in 2021. Our multiple stepwise regression results demonstrate that MBC is not a crucial factor affecting R_S (Supplementary Tables S1, S2). The reason for this phenomenon may be due to the decrease in the use efficiency of microorganisms. Microbial carbon use efficiency (CUE) refers to the distribution ratio between the MBC produced by organic matter catabolism and the C allocated by microorganisms for aerobic respiration (Schimel et al., 2022). The results of Manzoni et al. demonstrated that high temperature reduces the CUE of microorganisms (Manzoni et al., 2012) because the increase in T_S results in less C being allocated for microbial growth, which in turn reduces the ability of microbes to decompose organic resources to mitigate ecosystem C loss (Allison et al., 2010). During the growing season, T_S gradually increased, and microbes probably allocated more C for R_S than for MBC.

According to the results of the SEM, in addition to T_S, βG and PPO also significantly influenced R_S. Soil enzymes and microorganisms are involved in regulating the transformation of various organic matter and material circulation processes. Enzyme activity can be used as an indicator of microbial activity and plays a crucial role in decomposing organic compounds (Boerner R. E. J. et al., 2005). Previous studies have demonstrated that βG could participate in the decomposition of cellulose in litter (Caldwell, 2005). Litter is the most important source of organic matter input to the soil, and it can influence R_S by affecting the amount of labile C in the soil (Zhang et al., 2020). Therefore, βG has a significant positive effect on R_S. In addition, PPO can promote the accumulation of SOC by depolymerizing or aggregating lignin molecules and phenolic compounds in the soil (KIRK et al., 1987), thus positively influencing R_S.

Our results suggest that T_S, microbial biomass, and enzyme activity may be the main factors influencing soil microbial activity, which has important scientific implications for constructing microbial decomposition models that predict soil microbial activity under climate change in the future. To better understand the relationship between soil dynamics and C emissions, it will be necessary to incorporate climate data as well as R_S and microbial parameters into microbial decomposition models, which will be important for soil conservation and reducing soil C loss in the Loess Plateau.