Long-term Fertilization Enhances Soil Carbon Stability by Increase the Ratio of Passive Carbon: Evidence From Four Typical Croplands

Aims Soil organic carbon (SOC) plays an important role in improving soil quality, however how long-term fertilization inuences SOC and contrasting active carbon (AC) and passive C (PC) pools at large scale remains unclear. The aim of this study was to examine the effect of long-term fertilization on SOC, including AC and PC, across four typical croplands in China and to explore the potential relationships and mechanism. Methods We assessed the effect of different fertilization (standard and 1.5 × standard of inorganic fertilizer (NPK) with or without manure (M), with a control for comparison) at soil depths (0-20 cm, 20-40 cm, 40-60 cm) on SOC, AC and PC. Results We found that SOC, AC and PC increased in the order Control < NPK < NPKM < 1.5NPKM. 1.5NPKM resulted in a signicant increase in SOC, AC and PC, of 76.3%, 53.0% and 108.5% respectively across the soil prole (0-60 cm) compared with Control. The response ratio of PC to long-term fertilization was 2.1 times greater than that of AC across four sites on average. In addition, Clay was identied as the most important factor in explaining the response of AC and PC to different fertilization application, respectively. Conclusions that application carbon thus of cropland soil carbon

(2017), found that 11 years of chemical and manure fertilization can signi cantly increase SOC content. Majumder et al. (2008) also observed NPK plus organics increased SOC by 24.3% over the control in a 19year-old rice (Oryza sativa L.)-wheat (Triticum aestivum L.) cropping system in subtropical India, while the cropping with only NPK fertilization only maintained SOC content. However, in another study, no clear effects of 25 years of inorganic fertilization on the SOC storage were observed in a monoculture maize (Zea mays L.) cropping system in temperate north China because enhanced carbon input were offset by greater decay rates of soil C (Dou et al. 2016). Hence, understanding and characterizing the effect of fertilization on SOC accumulation is important for the development of sustainable long-term agriculture.
To better understand SOC accumulation dynamics under long term fertilization, total SOC can be separated as active (AC) and passive C (PC) pools according to stability (Chan et al. 2001). AC decomposes fast (Wu et al. 2012) and responds rapidly to changes in C supply before detectable change in total SOC, and is considered to be an early indicator of changes in soil quality (Zhang et al. 2018). PC on the other hand is stabilized against further microbial action due to physical protection and chemical recalcitrance. It dominates SOC stores, and can have a residence time of hundreds or thousands of years (Balesdent et al. 2018; Guo et al. 2017) It is, thus, critical for promoting SOC sequestration (Nath et al. 2018; Xu et al. 2018). Importantly, the in uence of organic versus inorganic fertilization on contrasting pools can be very different. Ding et al. (2012) found that 10 years of application of organic manure with chemical fertilization produced a greater size of PC and a decrease in AC, thus increased SOC stability, compared with the application of chemical fertilization alone. Huang et al. (2017), through a 6-year fertilization experiment found the opposite was true, and that AC increases were greater than for PC. Therefore, changes in SOC fractions under long-term fertilization still remain unclear. In this paper, we aimed to quantify the effect of different long-term fertilization practices on active and passive SOC fractions.
Previous studies investigating how fertilization affects SOC have mainly focused on the top 20 cm soil layer in cropland (Sun et al. 2013), which is the depth recommended by the IPCC (Penman et al. 2003 The experimental design comprised plots arranged in a randomized complete block design with three replicate blocks and each plot was isolated by 100 cm cement ba e plates. At GZL and ZZ, the plot sizes were 25 m×16 m, and at CQ and QY the plots were 10 m×12 m and 10 m×20 m, respectively. At each site, we focused on four treatments: Control (without fertilization), NPK (chemical fertilization), NPKM (combination of manure with NPK), 1.5NPKM (greater rate of manure with NPK). The N, P and K fertilizations were urea, calcium superphosphate, and potassium chloride, respectively. The C contained in urea is small compared with the manure inputs and urea in the soil tends to be rapidly mineralized to produce CO 2 . The source of amended manure is a mixture of cattle, horse and pig wastes. For the NPKM treatments, 30% of total N applied was inorganic, as urea, and the rest 70% was as organic manure. The rate of fertilization applied for 1.5NPKM was 1.5 times that of the NPKM, except at CQ which was 1.5 times that of the NPK. Manure was applied once a year before seeding of wheat at ZZ and CQ, and before seeding of maize at GZL. The manure was applied twice a year before seeding of wheat and maize at QY (Liang et al. 2019). The annual application rate of chemical N (applied as urea), P (applied as calcium superphosphate), K (applied as potassium chloride) fertilization and manure for various fertilization treatments are presented in Table 2. Table 2 Application rates of chemical fertilizer (kg ha − 1 ) and manure (Mg ha − 1 ) at the four long-term experimental sites in China The application amount of fertilization is N-P-K-M; N, P and K fertilizations were urea, calcium superphosphate, and potassium chloride, respectively; M, is a mixture of cattle, horse and pig wastes.

Soil sampling and processing
Soil samples were taken at 0-20, 20-40 and 40-60 cm depths after crop harvest in August 2010. Three soil samples were obtained randomly from each replicate plot using a 10-cm diameter auger. Then, three soil cores within each replicated plot were well mixed. Additional triplicate soil samples were taken from all the plots and depths using the ring method to determine the bulk density (Blake and Hartge 1986). Visible crop residues, root material and stones were removed during sieving. Soil was air-dried and passed through a 2 mm sieve for chemical analysis of soil organic carbon (SOC), total nitrogen (TN), AC and PC.
Soil analysis SOC and TN were determined by dry combustion using a C/N analyzer (EA-3000, EuroEA Elemental Analyzer). Gravimetric concentrations of SOC (g kg − 1 ) for each depth were converted to area-based stocks (Mg ha − 1 ) using the conventional approach: where stocks are in Mg ha − 1 , SOC concentrations are g kg − 1 , Bd is bulk density (Mg m − 3 ) and D is the depth interval of the soil layer (m).
The amount of SOC stored in the soil pro le (0-60 cm) was estimated as follows: Where SOC D is the SOC stock at 0-20, 20-40 and 40-60 cm three depths.
The AC and PC were determined through a modi ed Walkley-Black method ( The effects of fertilization on the variables (X) were quanti ed by the response ratio (RR) using the following equation: where X T and X C represent the mean of the fertilizer treatment and Control groups for variable X, respectively. The results are presented as the ratio of changes in the variables under fertilization. Positive percentage changes denote an increase due to fertilization application whereas negative values indicate a decrease in the respective variables.

Statistical analysis
To examine differences in the three target variables among the four fertilization treatments, data were analyzed using the Least Signi cant Difference (LSD) at 5% level of signi cance by using R "agricolae" package (Mendiburu 2020). Meanwhile, analysis of variance (ANOVA) was also used to evaluated the effects of site, fertilization treatments, soil depth, as well as their interaction on SOC, AC and PC. The treatments and blocking structures used were site * fertilization * depth and site. block / fertilization / depth, respectively. Second, to explore the changes of AC and PC under different fertilization treatments, we calculated the proportion differences and response ratios (RR) of AC and PC to fertilization additions.
Third, correlation analyses were then conducted to examine the relationships between RR and soil texture, environmental variables, and soil chemical properties by using "corrplot" package (Taiyun and Viliam 2017). Then, importance analysis (Random Forest) was used to explore the correlation factors controlling the RR of AC and PC by using the "randomForest" package. Finally, we used structural equation modeling (SEM) to explore the direct and indirect factors regulating RR, as well as to evaluate the contributions of these factors by assessing the degree of the standardized total effect (direct effect plus indirect effect) by using "sem" package.
We constructed SEM for the cropland in the consideration of potential differences in the mechanisms underlying RR. To obtain the nal SEM, the following two steps involving base model construction and model optimization were speci ed. First, we established a base model on the basis of empirical knowledge. Speci cally, based on the correlation and RF analysis between variables and RR, we included all variables that were signi cantly correlated with RR in the base model. We then built causal relationships between these variables and RR. In SEM, we assumed that soil chemical properties were expected to play direct roles in RR. Moreover, the RR is also indirectly and directly affected by the soil properties and environmental variables. Second, we optimized the base model on the basis of actual measurements. Speci cally, we rst examined all indices to ensure that no important paths were left out of the model, and then we removed paths with coe cients that were not signi cant at P < 0. The chi-squared (χ2) statistic, degrees of freedom (df), whole-model p-value, root mean square error of approximation (RMSEA), comparative t index (CFI) and standardized root mean square residual (SRMR) were used to assess the overall goodness of model t.  (Fig. 2). The PC was signi cantly increased by 76.7% and 108.5% under NPKM and 1.5NPKM treatments, but there was no signi cance difference under NPK (Fig. 3).
The fertilization treatments and soil depths as well as their interaction had signi cance effects on SOC stock, AC and PC (P < 0.001, Table S2). SOC stocks, AC and PC were generally decreased with soil depth at all treatments (0-20 cm > 20-40 cm > 40-60 cm) (Fig. 1, Fig. 2, Fig. 3 54.7%, respectively. We also found that the 1.5NPKM treatment resulted in PC representing more than 50% of SOC across the sites. Across the three sites, PC was greater than AC under 1.5NPKM treatments, expect GZL site.

Response ratio of SOC, AC and PC
The results show that long-term fertilization signi cantly increased the RR of SOC, AC and PC across all depths and sites (Fig. S2, Fig. 5 Modeling drivers of response ratio The RR was correlated with soil chemical properties (SOC, TN, C/N and pH), environmental variables (MAT and MAP) and soil texture (Clay) (Fig. S2). It shows that RR was positively with SOC (r = 0.32), TN (r = 0.31) and C/N (r = 0.16). However, it decreased with MAT (r = -0.23) and MAP (r = -0.14). In addition, it also shows that RR was negative with the Clay (r = -0.26).
SEM analysis showed that SOC, Clay, MAP and fertilization input had direct effects on RR. Together, these variables predicted 52% of variance in RR (Fig. 6). Speci cally, soil chemical properties, including SOC had direct positive effects on RR. Environmental variables, including MAT and MAP, had direct and indirect by mediating soil texture and SOC. Soil texture, including Clay, had direct effect on RR. Taking the indirect and direct effects together, Clay was the most important factors in uence the RR in cropland, followed by MAP, MAT, fertilization input and SOC (Fig. 7).

Effect of fertilization on SOC, AC and PC
The analysis revealed fertilization can signi cantly increase the SOC, AC and PC in the following order 1.5NPKM, NPKM and NPK (Fig. 1, Fig. 2, Fig. 3), when compared with Control. The greatest value under 1.5NPKM might be attributed to the greater rate of C addition than NPKM, NPK or Control as the 1.5NPKM treatment did not add more M at CQ, and there was no increase in SOC compared with NPKM at this site. In contrast, there were greater C inputs in the 1.5NPKM treatment and SOC increased at the other 3 sites (Fig. 1). Most research has reported that 1. We also found the signi cant increases in passive C in fertilizer addition in all depths (Fig. 2, Fig. 3). Effect of fertilization on response ratio of SOC, AC and PC In our study, the RR of PC to long-term fertilization was greater than AC at the four sites (Fig. 5), indicating a greater effect of fertilization on PC. There are many potential reasons for this. Although both AC and PC increased with fertilization input ( Fig. 2; Fig. 3), however, AC is the chief source of soil nutrients, and may be easily used by the microbes, so reducing AC remaining relative to PC in the soil. In contrast, PC is very stable under the protection of organic-mineral complexes and soil aggregates and is di cult to be used Driving factors on response ratio The SEM revealed that Clay was the major driver of the variation of RR in croplands (Fig. 7). Soil contains differently sized mineral particles, which are usually classi ed simply as sand, silt and clay (Ding et al. 2018). We found clay content has negative relationship with RR of soil fractions. Clay and silt particles contain mainly sesquioxides and layer silicates, and provide large speci c surface areas and numerous reactive sites at which carbon can be absorbed by strong ligand exchange and polyvalent cation bridges (Singh et al. 2017). Hence the response ratio of soil fraction may decrease with the Clay content because the Clay usually contain large amount of carbon, and thus the absolute changes in SOC are being divided by a larger number.
The results of the RF model showed Clay were identi ed as having the greatest importance as the main attributes to explain the response of AC and PC to fertilization addition, respectively (Fig. S5). Clay might increase the formation of organic-mineral complexes which can be accessible to soil Enzymes to produce active carbon in soil . Fertilization is essential for microbial growth and activities ( (Yang et al., 2018), thus allowing more microbial necromass accumulate and impacting the content of PC that returns to the soil directly. Secondly, fertilization can also stimulate the production of compounds in plants aboveground ) and root exudates which will increase the carbon input to the soil to accumulate more PC.

Conclusion
Long-term fertilization treatments increase SOC stock, AC and PC contents with the greatest values under 1.5NPKM across four typical agriculture sites. Greater applications of manure and chemical fertilization treatment play a key role in crop productively and future global C storage. The response ratio of PC to long-term fertilization was greater than AC at four sites indicating that fertilization increased PC more than AC which could enhance soil C stability. We found fertilization affected PC more than AC at all soil layers. The most important factors for explaining the response of AC and PC are Clay and Fertilization which indicated that the response of AC and PC were caused by two different kinds of mechanism. In short, these ndings increase our understanding of how long-term fertilization management affects SOC stock, AC, PC and their response ratio at different depths. And fertilization could enhance the stability of carbon.
Declarations Figure 1 Soil organic carbon stock across four sites in different soil depth under various fertilizations. See Table 1 and Table 2   Active carbon across four sites in different soil depth under various fertilizations. See Table 1 and Table 2 for experimental treatments in detail. The lower-case letters over the bar indicate signi cant differences among different fertilization treatments at P < 0.05 level of probability. Error bar in Ave (n = 4) and GZL, ZZ, CQ, QY (n = 9) indicates standard error of the mean of treatments for the individual sampling sites with respect to the individual soil depth. Passive carbon across four sites in different soil depth under various fertilizations. See Table 1 and Table   2 for experimental treatments in detail. The lower-case letters over the bar indicate signi cant differences among different fertilization treatments at P < 0.05 level of probability. Error bar in Ave (n =4) and GZL, ZZ, CQ, QY (n = 9) indicates standard error of the mean of treatments for the individual sampling sites with respect to the individual soil depth.  Proportion of active and passive carbon pools of soil organic carbon under different fertilizations. See Table 1 and Table 2 for experimental sites and treatments in detail. Error bar in Ave (n = 4) and GZL, ZZ, CQ, QY (n = 9) indicates standard error of the mean. Colors correspond to different soil pools: AC (red), PC (blue). Response ratio (to Control) of active and passive carbon under different fertilization at three depths and four sites as well as an average. See Table 1 and Table 2