Multiple trade‐offs between maximizing yield and minimizing greenhouse gas production in Chinese rice croplands

Globally, paddy fields are a major anthropogenic source of greenhouse gas (GHG) emissions from agriculture. There is, however, limited understanding of relationships between GHG production with fertilizer management, rice varieties, and soil variables. This information is crucial for minimizing the climatic impacts of rice agriculture. Here, we examined the relationships between soil GHG production and management practices throughout China. The current doses of N‐fertilizer (73–272 kg ha−1) were negatively correlated with rice yield and with CO2 or CH4 production and positively correlated with N2O production, thus suggesting N‐overfertilization. Impacts on soil traits such as decreasing pH or the availabilities of other nutrients could be underlying these relationships. Rice yield was highest, and GHG production was lowest at sites using intermediate levels of P‐ and K‐fertilization. CO2 and CH4 production and emissions were positively related with soil water content. The yield was higher, and N2O productions were lower at the sites with japonica rice. Our results strongly suggest that current high doses of N‐fertilizers could be reduced to thus avoid the negative effects of excessive N input on GHG production without any immediate risk of rice production loss. Current intermediate doses of P‐ and K‐fertilization should be adopted across China to further improve rice production without the risk of GHG emissions. The use of different rice varieties and strategies of water management should be reexamined in relation to crop production and GHG mitigation.

the growing population (Food and Agricultural Organization of the United Nations, 2009). Sustaining soil fertility and increasing rice yields are therefore of utmost importance. An increased nutrient supply can not only stimulate the growth and grain yield of rice plants (Ali, Oh, & Kim, 2008;Wang et al., 2014) but can also influence the potential of paddy fields to produce and emit greenhouse gases (GHGs). Globally, paddy fields are a major anthropogenic source of GHG emissions from agriculture (Tan, 2011). Paddy fields are very important sources of GHG, especially methane (CH 4 ) and nitrous oxide (N 2 O; Myhre et al., 2013), so minimizing the release of these very potent GHG could contribute to mitigating their adverse impacts on climate change (Li, Salas, Deangelo, & Rose, 2006). Chinese GHG emissions from agricultural systems account for~40% of Chinese GHG emissions, hence requiring detailed investigations.
Furthermore, because 60% of the Chinese population depend on rice-based food, so protecting China's rice production for food security is important (Zhu, 2006).
There is, however, limited understanding of relationships between GHG production with fertilizer management, rice varieties, and soil variables. This information is crucial for minimizing the climatic impacts of rice agriculture, especially for the agricultural sustainable development in China. Improving the status of soil nutrients in paddy fields for improved rice yield while decreasing GHG emissions, or at least not increasing it, is a challenging option.
However, intense fertilization can also induce rises in GHG emissions in paddy soils , and great soil nutrient concentration also can be related to GHG emissions in paddy soils . Rice crops in China are frequently overfertilized (Cheng & Li, 2007), so a general analysis of the relationships of fertilization and soil traits with GHG emissions and crop yield is necessary to detect the level of overfertilization, GHG emissions and yield, and moreover, to improve the future management of the sustainable development of rice agriculture. Here, we examined the relationships between soil GHG production and management practices throughout China.
We hypothesized that different fertilization practices, strategies of rice-crop management, rice varieties, concentrations of soil carbon (C) and other nutrients, salinity, or/and pH would explain a large part of the differences in GHG emissions and yield among sites, especially, the fertilization amount increment will increase the GHG emissions but not the yield. Our results will provide information for improving strategies and managing soil conditions toward more favorable traits to avoid a possible increase in GHG production without yield loss. We pursued this objective by determining (a) the relationships among fertilization dose, soil GHG production, and rice yield in China and (b) the relationships among GHG production, rice variety, and environmental traits, such as region (sites), location (coastal vs. inland), and cropping system (single or double) on these relationships.

| Study area
This study was conducted throughout China. China has 2.45 × 10 7 ha of cultivated rice, and 90% of the paddies are in the subtropics. China has a large area of paddy rice, diverse soil types, and different tillage and fertilizer management practices, all of which may affect the content and distribution of nutrients and GHG emissions. We collected the soils in the whole China rice cultivation areas choosing sites with contrasting fertilization management. The studied ranges of fertilization were 73-272 kg ha −1 for N fertilizer, 48-150 kg ha −1 for P 2 O 5 , and 45-270 kg ha −1 for K 2 O. The main characteristics of the different studied sites are shown in Table S1. To analyze the role of sea proximity, we separated the studied sites between inland and coastal rice crops depending on the distance to sea line; when less than 20 km was considered coastal paddy field, and more than 20 km was considered inland paddy field. To analyze the role of fertilization intensity,  (Sun, Huang, Zhang, & Yu, 2010). Moreover, the sampling site characteristics and paddy field management were also investigated. The collected soil samples were also used for all analyses. The samples were air-dried, and roots and visible plant debris were removed. Total C and N contents were measured using a Vario EL III Elemental Analyzer (Elementar Scientific Instruments, Hanau, Germany, . Labile organic carbon (LOC) content was determined by digestion with 333 mM KMnO 4 (Wang, Lai, Wang, Pan, & Zeng, 2015), whereas NH 4 + and NO 3 − were determined by extracting the soils with 2 M KCl (Lu, 1999). Available N content was the sum of NH 4 + and NO 3 − . The total P content was determined, first by perchloric-acid digestion and then using a sequence flow analyzer (San++, SKALAR Corporation production, Breda, the Netherlands). To determine soil P availability, we used the Mehlich extraction method and then using a sequence flow analyzer (San++, SKALAR Corporation production, Breda, the Netherlands).
The soil CO 2 , CH 4 , and N 2 O productions were determined using anaerobic incubation consistent with soils under water saturation . Thirty grams of fresh soil of the five core samples for each site were placed in 120-mL incubation bottles, and two volumes of distilled water were added. The bottles were purged with N 2 for 2 min to replace the O 2 and were then sealed with a rubber stopper and incubated at 25 C for 3 days. Five milliliters of gas were extracted from the headspaces each day, about 24-hr interval in four times during incubation: 0, 24, 48, and 72 hours during incubation experiments. This method has been successfully used in several previous studies Wassmann et al., 1998). We also use the method of Xu, Chen, and Xiong (2016) to calculate annual gas emissions from some days (in our case 3 days) of sampling . used as a carrier gas (30 ml min −1 ), and a make-up gas (95% argon and 5% CH 4 ) was used for the electron capture detector. The gas chromatograph was calibrated before and after each set of measurements using 503, 1,030, and 2,980 μl CO 2 L −1 in He; 1.01, 7.99, and 50.5 μl CH 4 L −1 in He and 0.2, 0.6, and 1.0 μl N 2 O L −1 in He (CRM/RM information centre of China) as primary standards . We used linear equations for calculating CO 2 , CH 4 , and N 2 O productions.
Other soil variables were also analyzed. Bulk density was measured using three 15 × 3 cm cores  and was estimated by core mass dry weight divided by core volume, and represent the averaged bulk density of 0-15 cm. Soil water content was measured by the drying method (Lu, 1999

| Determination of soil C and nutrient contents
The total C, N, and P contents, and LOC, available N, available P, NH 4 + -N, and NO 3 − -N contents in the 0-15 cm soil profile were estimated by following the approach of Mishra, Ussiri, and Lal (2010): where C S is the total C, N, or P contents or LOC, available N, or P content or LOC, available N, available P, NH 4 + -N or NO 3 − -N content (g kg −1 ), ρ b is the bulk density (kg m −3 ), D is the thickness of each soil layer (0.15 m).
The total C, N, and P contents, and LOC, available N, available P, NH 4 + -N and NO 3 − -N contents were calculated for each region by multiplying area content and areas of each paddy distribution. The contents for each measured variables across China were calculated as the sum for all regions.

| Determination of soil CO 2 , CH 4 , and N 2 O productions
We used linear regressions for calculating soil CO 2 , CH 4 , and N 2 O productions (Wassmann et al., 1998): where P is the rate of CO 2 , CH 4 , or N 2 O production (μg −1 g −1 day −1 ), dc/dt is the recorded change in the mixing ratio of CO 2 , CH 4 , and N 2 O in the headspace over time (mmol mol −1 day −1 ), V H is the volume of the headspace (L), Ws is the dry weight of the soil (g), MW is the molecular weight of CO 2 , CH 4 , or N 2 O (g), MV is the molecular volume (L), T is the temperature (K), and T st is the standard temperature (K).
The potential productions of CO 2 , CH 4 , and N

| Statistical analyses
One-way ANOVAs with N-, P-, and K-fertilization doses categorical (low, intermediate, and high intensity), regions, inland-coastal environments and cropping systems as independent variables and gas production variables, and yield as a dependent continuous variable were conducted with Bonferroni post hoc tests. The data were checked for normality and homogeneity of variance, and if necessary, were logtransformed. We used the Benjamini-Hochberg procedure to control the rate of false discovery (Benjamini & Hochberg, 1995) to analyze the relationships between gas emissions and all the studied environmental soil and climate variables. The different individual analyses were listed in rank order according with their ascending p value.
Thereafter, for each single analyses, each p value was divided by the total number of test and thereafter multiplied by the false discovery rate (habitually 0.25), then the values below .05 were considered as significant. We also used Tukey's method (Tukey, 1977) to detect and remove outliers.
We also performed multivariate statistical analyses. We used principal component analyses (PCAs) to determine the overall differences of the soil variables and CO 2 , CH 4 , and N 2 O productions rates and annual accumulated emissions among fertilizer doses. We used all variables given the scarce multicollinearity existing among variables (see Table S1), with no R 2 > 0.6 between any pairwise variables. We have also estimated VIF for each independent fixed variables in the mixed models. We conducted one-way ANOVAs with Bonferroni post hoc tests of the scores of the first PC axis to determine differences among the treatments. We then used general discriminant analyses (GDAs) to determine the overall differences of soil traits; CO 2 , CH 4 , and N 2 O productions rates and annual accumulated emissions and yield among fertilizer doses and productivities among sites with different yields. Discriminant analyses consist of a supervised statistical algorithm that derives an optimal separation between groups established a priori by maximizing between-group variance while minimizing within-group variance (Raamsdonk et al., 2001 Hochberg procedure to control for rates of false discovery (Benjamini & Hochberg, 1995).
We used structural equation modeling (SEM) to study the total effects of fertilizer doses on accumulated gas emissions by both direct and indirect effects of soil traits. We fit the different models using the SEM R package (Fox, Nie, & Byrnes, 2013) and determined the minimum adequate model using AIC. Standard errors and significance levels (p values) of the total, direct, and indirect effects were calculated using bootstrapping (1,200 repetitions; Davison, Hinkley, & Schechtman, 1986;Mitchell-Olds, 1986).

| RESULTS
3.1 | Effects of fertilization doses, regions, rice variety, inland-coastal environments, and cropping systems on soil GHG productions and rice yield One-way ANOVAs with N-, P-, and K-fertilization doses as independent categorical (low, intermediate, and high intensity) variables and gas production variables, and yield as a dependent continuous variable, indicated that the lowest levels of N fertilization (<100 kg N ha −1 yr −1 ) were associated with the highest annual yields. However, the highest levels of N fertilization (>150 kg N ha −1 yr −1 ) were associated with the lowest annual yields (Figures 1d and S1). The lowest doses of N fertilizer were associated with the highest soil CO 2 and CH 4 productions and the lowest N 2 O productions, with the opposite patterns at the highest doses of N fertilizer. The intermediate doses of P (60-75 kg P 2 O 5 ha −1 yr −1 ) and K (60-70 kg K 2 O ha −1 yr −1 ) fertilizers were the best, because they were associated with the highest yields and the lowest CO 2 and CH 4 productions (Figures 2a,b, S2, 3a, b, and S3). The correlations among all the studied variables were shown in Table S2.
The balance between yield and GHG production was worst in the East China rice crops, with the highest productions of CO 2 , CH 4 , and N 2 O and the lowest annual yield (Figures 4 and S4). Coastal rice crops have higher CO 2 and CH 4 production (Figures 5a,b and S5a,b) and lower yield than inland rice crops (Figure 5d). Total annual yield was notably similar at the sites with one and two annual rice crops ( Figure 6d). The CO 2 production rates were higher at sites with one annual crop than at sites with two annual crops ( Figure S6a,c).
We excluded hybrid and glutinous rice varieties from the analysis of the effects of rice variety, because both were only at one site each.
We focused on japonica and Hsien rice varieties, which were the main varieties planted at the sites. On average, the yield was higher, and N 2 O productions were lower at the sites with japonica rice ( Figure S7).

| Gas productions and environmental traits
We observed scarce relationships with studied gas emissions and production with the studied soil variables, the most strong was the positive relationships between total soil N concentrations and CO 2 emission and production and between soil water content and methane emission and production (  Coastal rice crops have higher CO 2 and CH 4 emissions and lower yield than inland rice crops. This was associated with higher soil N, water, and clay contents in coastal rice crops (Figure 7c). Figure 7d shows the overall differences in the environmental variables between the sites with one and two annual crops. These two groups of sites were clearly separated along the PC1 axis, with single-crop sites plotted toward higher annual yields and soil P contents, salinities, C:N ratios and pHs, whereas the double-crop sites were plotted toward higher soil N, water and LOC contents and N:P ratios, and higher CO 2 productions.
The best mixed models (based on a low AIC and the highest R 2 and parsimoniousity), with rice variety and environmental traits as fixed independent variables; location, topography, site, and plot as random factors; and GHG productions as dependent variables, indicated that 26% of the total variance of annual accumulated CO 2 productions was explained by the length of the growth period, soil water content, and bulk density (Table S3). Thirty-nine percent of the total variance of CO 2 productions rates was explained by the growthperiod length, soil water content, and P-fertilizer dose. Forty-seven and 50% of the total variance of annual accumulated CH 4 productions and CH 4 productions rate, respectively, were explained by the F I G U R E 3 Rice yield and total accumulated productions of CO 2 , CH 4 and N 2 O for the K-fertilizer annual doses (low <70 kg ha −1 , intermediate 71-81 kg ha −1 , high more than 81 kg ha −1 ). Different letters indicate significant statistical differences (p < .05) F I G U R E 4 Rice yield and total accumulated productions of CO 2 , CH 4 , and N 2 O in the various regions. Different letters indicate significant statistical differences (p < .05) growth-period length, soil water content, water source (river or groundwater), rice variety, bulk density, total soil N content, and P-fertilizer dose. Nineteen and 21% of the total variance of annual accumulated N 2 O productions and N 2 O productions rate, respectively, were explained by the cropping systems, soil P availability, and soil NO 3 − -N content. All six mixed models (Table S3), corresponding to each of the production variables, explained >99% of the corresponding total variance, while taking into account the random variables.

| Relationships between annual yield and gas productions; fertilization effects
The GDA indicated that the crop sites with high yields generally had the highest GHG productions, whereas crop sites with moderate yields generally had the lowest GHG productions ( Figure S8). Lower doses of N fertilizer were surprisingly associated not only with higher yields but also with higher CH 4 productions ( Figure S9 When the productivities at the various doses of N, P, and K fertilizers (yield kg −1 fertilizer) were used as grouping dependent factors in the GDA, the sites with high N productivity were associated with the highest N 2 O productions, sites with intermediate N productivity were associated with the highest CH 4 productions, and sites with low N productivity were associated with higher CO 2 productions ( Figure S12). The lowest CO 2 , CH 4 and N 2 O productions F I G U R E 5 Rice yield and total accumulated productions of CO 2 , CH 4 , and N 2 O at inland vs coastal sites. Different letters indicate significant statistical differences (p < .05) F I G U R E 6 Rice yield and total accumulated productions of CO 2 , CH 4 , and N 2 O at sites with one vs two rice crops. Different letters indicate significant statistical differences (p < .05). Different letters between brackets indicate marginal significant statistical differences (p < .1) were associated with sites with intermediate P fertilization ( Figure S13), and sites with intermediate K fertilization were associated with the lowest CO 2 and CH 4 but not N 2 O productions ( Figure S14).

| SEM models
The SEM provided further evidence of the complex relationships among fertilizer doses, GHG productions, and annual yield. The N fertilization had a positive direct effect on GHG productions and a negative effect on annual yield, and P fertilization had the opposite pattern ( Figure 8), and K fertilization did not significantly affect GHG productions but generally negatively affected yield (Figure 8c). The N fertilization also had indirect negative effects on annual yield by increasing the amount of labile soil carbon content and by lowering soil pH. We also detected a negative direct and indirect effect, by decreasing soil water content, effect of bulk density on soil CO 2 productions ( Figure 8b).

| DISCUSSION
We studied potential emissions of carbon dioxide (CO 2 ), CH 4 , and Different studies have reported contradictory effects of N fertilization on GHG emissions from rice crops (see the review by Cai, Shan, & Xu, 2007). Several studies, however, have observed an increase in N 2 O emissions with increases in N-fertilizer dose in rice croplands Zhang et al., 2014).
Our data analyses indicated that P-and K-fertilizer doses generally had positive effects on rice yield and decreased GHG productions.
The relationships among P-and mainly K-fertilization, yield, and GHG productions in rice croplands have not been studied much, as compared with the corresponding relationships with N-fertilization. Our results are nonetheless consistent with other experimental studies (Datta, Santra, & Adhya, 2013;Li, Wang, et al., 2013;. Datta et al. (2013) found that N fertilization explained more of the changes in CH 4 emissions than P or K fertilization. In contrast, Li, Wang, et al. (2013) and  observed that increasing the dose of P fertilizer decreased CH 4 and N 2 O emissions but increased yield.
However, our study has some potential limitations. The incubation period of 3 days could not be sufficient to capture the overall F I G U R E 8 Structural equation modeling (SEM) models with soil CO 2 production rate (a), soil accumulate CO 2 production (b) and rice crop yield (c) as end endogen variables. Numbers on the arrows are the estimates of the effects of one variable (near the beginning of the arrow) over another (near the end of the arrow) and the corresponding p values (in parentheses). Red arrows indicate negative relationships, and black arrows indicate positive relationships [Colour figure can be viewed at wileyonlinelibrary.com] patterns of GHG production and emission, and thus has limited reflection of the field condition, but it is still very useful to compare the potential emission among studied sites. In the paddy fields, the first peak of methane emission generally occurs within a month after transplanting, just according with our study. But a second peak would occur at approximately 2 months, and this is mainly governed by the stable low soil redox potential and neutral soil pH, and the increased release of plant-borne carbon sources (e.g., Ly, Jensen, Bruun, & de Neergaard, 2013;Vu et al., 2015). This can be partially corrected by using Xu et al. (2016) method to estimate all year gas emissions. But all in all, this can explain why the estimated methane production in this paper is lower than in other reports in Asian countries (Vo et al., 2018;Yan, Ohara, & Akimoto, 2003).
Our results suggest that rice yield can be increased without increasing or even decreasing GHG productions. For example, our results indicated that yields were higher for the japonica than the Hsien varieties. The N 2 O productions were several times higher for the Hsien varieties, leading to a better balance between yield and GHG productions in fields with japonica varieties, even though CH 4 productions were clearly higher for the japonica varieties. Adequate irrigation is fundamental for assuring high rice yield (Sun et al., 2016), but our data also suggest that GHG productions can be reduced by regulating doses of N-, P-, and/or K-fertilization. We have also observed that CH 4 and N 2 O productions were much higher at sites irrigated and flooded with river water than at sites irrigated and flooded with groundwater or water from superficial reservoirs, whereas average yield did not differ significantly between these sites.
This result is difficult to interpret in the context of this study; further studies should aim to find out the cause of these differences. The balance between yield and GHG production was worst in the East China rice crops, with the highest productions of CO 2 , CH 4 , and N 2 O because in this area the temperature is relatively higher, and there are more active substrates . The North China and Northwest China rice crops had the opposite patterns because in this area, the lower temperature limits the substrates decomposition, such as soil organic carbon, and then the microbes act to convert plant residues into humus in the soil (Cui et al., 2008). The lower temperature can decrease decomposition and the CO 2 and CH 4 release from soil by mediating the microbe growth (Tang, Cheng, & Fang, 2017).
Moreover, in these areas paddy soils had relative higher pH and water comes from rivers that can provide more substrate to the paddy and also have longer growth period and more illumination than in other areas of china. Furthermore, the North China and Northwest China rice crops lower GHGs production was related with the higher C:N ratio. The C:N ratio controls the CO 2 and CH 4 release. There is more limited carbon decomposition with relatively higher C:N ratios (Windham, 2001). The soil GHG productions tended to be lower (significantly for CO 2 ) at inland than coastal sites because he coastal paddy fields had higher carbon and nitrogen concentration, and therefore more substrates for soil GHG productions (Delaune, White, Elsey-Quirk, Roberts, & Wang, 2018).
Moreover, the CO 2 production rates were higher at sites with one annual crop than at sites with two annual crops because the two annual areas, mostly in the south of China, had soils rich in ferric oxide, which, in turn, favored C fixation in soil, and more stable soil carbon, thus decreasing C release in form of methane (Wang et al., 2014). Furthermore, CH 4 production was higher under single crop.
Single crop areas had lower temperature during the whole year, thus lowering decomposition, increasing carbon storage in the soil, and providing more substrates for CH 4 production when the temperature increases during rice growth period. However, N 2 O is higher under double crops because of more applications of N fertilizer.
CH 4 production was higher under single crop. Single crop areas had lower temperature during the whole year, thus lower decomposition, and more carbon can be storage in the soil and more substrates is able for CH 4 production when the temperature increase during rice growth period. However, N 2 O is higher under double crops, because along the year the N fertilizer applied is in higher amount in the doubles crop areas.
Our multivariate analysis of all data from all sites indicated that the highest average yields were accompanied by the highest GHG productions. Optimizing rice production and GHG production is thus interesting and challenging. Our analyses provide some clues for optimization, suggesting that some advances could be achieved by adjusting N-fertilization to a level that improve rice production but avoid the negative effects of excessive N input. Further, GHG productions could be reduced by maintaining adequate levels of P and K fertilizers, mostly at intermediate doses of the current range of P and K fertilization across China, and by reexamining the use of different rice varieties and strategies of water management.

| CONCLUSIONS
Rice production has historically been improved using N fertilizers, but we found that the current paddy field sites in China with relatively low N fertilization had high rice production and low soil CO 2 and CH 4 potential productions, even though rice yield was positively correlated with soil N availability. In contrast to N fertilization, sites using intermediate doses of P and K fertilizers had the highest rice yields and the lowest soil GHG potential productions. The large capacity of soil to accumulate P in nonavailable forms and the large capacity of K leaching by the high use of water in rice crops together with a less direct role of P and K in CH 4 and N 2 O productions could explain these results.
The analysis of all our data strongly suggests that increasing rice yield and minimizing GHG productions can be further optimized in