Agriculture is a major anthropogenic source of GHGs, and climate-change concerns have spurred efforts to reduce GHG emissions from agricultural systems11. Rice is a staple food for nearly half of the people in the world12. Rice paddies account for 11% of the planet’s arable land12 and are responsible for 10% of human-induced CH 4 emissions5, or 20% of total agricultural CH 4 emissions5. The global warming potential of GHG emissions from rice systems is roughly four times higher than either wheat or maize13.

World cereal production must increase by 1.3% yr−1 until 2025 to meet the food demands of a growing world population7, and stabilizing climate requires reducing GHG emissions associated with agricultural production. Thus, efforts to achieve both goals must account for potential trade-offs between them, suggesting the utility of metrics that incorporate both, such as GHG emissions per unit yield (rather than per unit land area)8,9,10. However, little is known about the response of yield-scaled emissions to environmental factors beyond farmers’ direct control, such as atmospheric CO 2 concentrations and climate.

Through fossil-fuel consumption and land-use change, humans are rapidly increasing the amount of CO 2 in the atmosphere and warming the planet14. Although increased atmospheric CO 2 is known to stimulate both rice yield and CH 4 emissions from rice paddies1,3, its combined effect on yield-scaled CH 4 emissions is unclear. Increased temperatures have been found to suppress rice yields4, but studies on the effects of warming on CH 4 emissions have so far been inconclusive.

Heterogeneity in studies of GHG emissions (in terms of system studied as well as experimental design and results) makes it difficult to determine their global response to elevated CO 2 or warming treatments from individual observations. This problem can be overcome by a quantitative synthesis of results across multiple studies. Thus, we used meta-analysis15 to summarize the effect of atmospheric CO 2 enrichment and warming on yield, CH 4 fluxes, yield-scaled CH 4 fluxes and root biomass from rice paddies, using observations from 63 published studies (see Supplementary Table S1 and Data S1–S8 and Notes). As responses of yield and CH 4 emission were correlated with the degree of experimental warming but not with the degree of CO 2 enrichment (see Supplementary Figs S1–S4), we normalized treatment effects in warming studies for the degree of temperature increase. All observations were analysed using two different weighting functions (see Methods).

Overall, increased CO 2 stimulated CH 4 emissions by 42.2% and rice yields by 24.6% (Fig. 1a and Supplementary Table S2). These results are similar to the average CO 2 responses of +43.4% for CH 4 emissions and +23% for rice yield that were reported in previous meta-analyses1,3. In studies that measured both CH 4 emissions and rice yield, increased CO 2 stimulated yield-scaled CH 4 emissions by 31.4%. Warming significantly decreased yield by 14.6% per 1 °C, but we found no evidence that warming affected CH 4 emissions (Fig. 1b and Supplementary Table S3). In studies in which both CH 4 emissions and yield were measured, warming caused a significant increase in yield-scaled CH 4 emissions of 11.8% per 1 °C.

Figure 1: Results of a meta-analysis on the response of CH 4 emissions, yield, yield-scaled CH 4 emissions and root biomass from rice paddies to increased levels of atmospheric CO 2 and warming. a, Effects of increased CO 2 . Results are based on 27, 184, 23 and 67 observations for CH 4 emissions, yield, yield-scaled CH 4 emissions and root biomass respectively. b, Effects of warming. Results are based on 42, 109, 13 and 82 observations for CH 4 emissions, yield, yield-scaled CH 4 emissions and root biomass respectively. Effect sizes in all meta-analyses were weighted by replication. Error bars indicate 95% confidence intervals. Full size image

Climate models featured in the fourth Intergovernmental Panel on Climate Change assessment estimate that by the end of the twenty-first century, land temperatures in rice-growing regions will roughly be 4 °C warmer than today16 (see Supplementary Methods). To estimate the combined effect of 4 °C warming and increased atmospheric CO 2 , we first determined whether CO 2 enrichment and warming showed interactive effects. The results of individual studies varied in this respect. For instance, of the five studies that assessed the effect of increased CO 2 and warming on CH 4 emissions, two studies found significant CO 2 ×warming interactions (Supplementary Data S9). However, one of these studies showed a positive interaction (that is, effects of increased CO 2 were stronger at high temperatures, and warming effects were stronger at increased CO 2 ), whereas the other study showed a negative interaction. Thus, we used meta-analysis to test whether CO 2 enrichment and warming showed interactive effects, using results from all factorial CO 2 ×warming experiments (see Methods and Supplementary Data S9–S12). As we found no significant interactions between warming and CO 2 (Supplementary Fig. S5), we calculated the combined effect of CO 2 enrichment and warming by multiplying their respective average treatment effects. Using this approach, we estimate that 4 °C warming in combination with CO 2 enrichment will increase CH 4 emissions by 58.0%, decrease rice yields by 33.8% and increase yield-scaled CH 4 emissions by 105.5%. Combined CO 2 ×warming effects for a range of warming predictions for rice-growing regions can be found in Supplementary Table S4.

The estimates of the response of CH 4 emissions and yield-scaled CH 4 emissions are likely to be low for two reasons. First, in factorial experiments, the observed combined effect of warming and CO 2 enrichment on CH 4 emissions was 25% higher than expected based on the product of their single-factor effects (Supplementary Fig. S5). Similarly, the observed combined effect of warming and CO 2 enrichment on yield-scaled CH 4 emissions was 29% higher than expected. Although not significant, these results suggest that our estimates for combined effects on CH 4 emissions and yield-scaled CH 4 emissions are probably about 25–29% too low.

Second, the negative effect of warming on rice yield (and consequently the positive effect on yield-scaled CH 4 emissions) increased with the control temperature of warming studies (Fig. 2), suggesting that the effects of warming will intensify as global temperatures continue to rise. The negative relationship between control temperature and warming effects is consistent with model predictions that climate warming will affect rice yields in temperate regions less strongly than in tropical regions17.

Figure 2: Effects of warming on rice yield and yield-scaled CH 4 emissions versus the control temperatures in warming experiments. a, The normalized effect of warming on rice yield (ln R TN ) versus the control temperature in the warming experiment. ln R TN is significantly correlated with the control temperature (r2 = 0.30,p<0.01). b, The normalized effect of warming on yield-scaled CH 4 emissions (ln R TN ) versus the control temperature. ln R TN is significantly correlated with the control temperature (r2 = 0.45, p = 0.01). Full size image

Why does CO 2 enrichment increase the amount of CH 4 emitted per unit yield? As methanogenic Archaea rely on C input by plants as their ultimate source of organic substrates18, root responses to CO 2 enrichment partly determine the response of CH 4 emissions1. Increased CO 2 stimulated root biomass to a larger degree than yield (47.8% versus 24.6%, Fig. 1a and Supplementary Table S2). Thus, our results suggest that with increasing CO 2 , rice plants allocate a larger part of assimilated C to roots than under ambient CO 2 , stimulating soil C input and CH 4 emissions to a larger degree than yields.

The production of CH 4 is determined both by the activity of methanogenic Archaea and by the availability of substrate18. Microbial activity such as CH 4 production is generally stimulated by temperature until an optimum is reached, after which it decreases owing to inactivation of the enzymes involved19. Indeed, production of CH 4 in four different anaerobic paddy soils was found to increase with temperature up to around 32–41 °C (ref. 20), after which it declined. The average air temperature applied in the heated treatments of warming studies on CH 4 emissions in our analysis ranged between 23.5 and 34.3 °C, suggesting that the effect of warming on methanogenic activity was positive in experiments with relatively low ambient temperatures, but smaller or negative under relatively warm ambient temperatures. Warming effects on root growth were positive but small compared with CO 2 responses (Fig. 1b and Supplementary Table S3), consistent with a positive, but small and non-significant effect of warming on CH 4 emissions. The response of yield-scaled CH 4 emissions to warming was mostly determined by the negative response of yield to warming, which was almost universal across studies (Supplementary Data S4). Declines in rice yield due to warming have been explained by an increase in floral sterility21 and by increased night-time temperatures suppressing assimilate accumulation4,22.

Besides CH 4 , nitrous oxide (N 2 O) is an important GHG emitted from rice paddies. However, N 2 O emissions on average account for only 11% of the global warming potential of GHG emissions from rice paddies13. Moreover, the only study so far that measured the response of N 2 O fluxes from rice paddies to increased CO 2 found no significant treatment effects23, further suggesting that N 2 O will play a minor role in changing the GHG intensity of rice cultivation under future atmospheric conditions.

Global climate and atmospheric changes have previously been suggested to increase GHG emissions and yields from rice cropping systems. For instance, a rise in temperature of 2 °C has been predicted to increase CH 4 emissions from rice paddies by 20% (ref. 24). Field data from an observational study suggest that rice yields decrease by 15% per degree average growing season temperature (and 10% per growing season minimum temperature)4, similar to our findings. However, this value has recently been revised to a smaller decline25. As yield-temperature relationships based on observational studies have not yet been synthesized, a full comparison with results from manipulative experiments is not yet possible.

The meta-analytic approach used here, synthesizing results across 63 studies, suggests that future CH 4 emissions from rice systems will mostly increase owing to increasing atmospheric CO 2 and less owing to warming, but that both factors contribute to an increased GHG intensity of rice cultivation.

Compared with other cereals, rice production systems show a large potential for reduction in yield-scaled CH 4 emissions through management practices13. A recent meta-analysis suggests that mid-season drainage can reduce CH 4 emission from rice paddies by as much as 52–60% (ref. 26), whereas replacing urea as a fertilizer with ammonium sulphate reduces CH 4 emissions by 40% (ref. 27). It should be noted that no study so far assessed whether mitigation efforts and climate change interact to influence CH 4 emissions from rice paddies. Without such interactions, mitigation practices may become more attractive over time, as similar percentage reductions will result in larger absolute reductions of CH 4 emissions. Clearly, the interaction between climate change and mitigation efforts on CH 4 emissions presents a knowledge gap that needs to be addressed. Regardless, our finding that future atmospheric conditions increase CH 4 emissions from rice production stresses the need to search for mitigation practices to curb GHG emissions.

Mid-season drainage and using alternative fertilizers (for example, ammonium sulphate instead of urea) may be cost-efficient means to reduce future GHG emissions from rice cultivation28. It is estimated that in 2020, at a break-even price (that is, the C price at which mitigation practices become economically viable) of US$10 per tonne CO 2 equivalents (CO 2 -eq.), global GHG emissions from rice cultivation can be reduced by 20% with little or no loss in yield28. A similar percentage reduction in GHG emissions from upland cropping systems would be achieved only at break-even prices about twice as high28, and is unlikely to be achieved for livestock management at any cost28. The future trajectory of C prices is highly uncertain; assuming a climate target of 650 ppmV CO 2 -eq., predicted C prices range between US$1 and US$35 depending on the nature of international participation in emission mitigation29. However, even at a break-even price of US$0 per tonne CO 2 -eq. (that is, at no extra cost to rice farmers), GHG emissions from rice cultivation can still be reduced by 10% (ref. 28).