Current yield gaps in rice cropping systems. Across cropping systems, the number of rice crops grown on the same piece of land during a 12-month period can range from one in non-tropical regions to three in tropical environments (Supplementary Figs. 1 and 2). Here we report metrics on a per-crop basis by averaging values across the rice crops within each cropping system where more than one rice crop is grown each year. Similarly, average values reported in this study are weighted according to the annual rice harvested area in each cropping system. At a global scale, yield potential averaged 9.4 Mg ha− 1 crop− 1, ranging from 5.9 to 14.8 Mg ha− 1 crop− 1 across the 32 rice cropping systems included in our analysis (Supplementary Fig. 3A). Average yield potential is higher in non-tropical regions than in tropical regions (9.8 versus 9.0 Mg ha− 1 crop− 1). Lower productivity per crop of tropical rice is more than compensated by higher cropping intensity as tropical regions have longer growing seasons that allow up to three rice crops each year in the same field (Supplementary Figs. 1 and 2). As a result, rice systems in tropical areas have greater annual potential productivity than in non-tropical regions (18.3 versus 13.6 Mg ha− 1 year− 1) (Supplementary Fig. 3A). In our study, all rainfed cropping systems are in lowland environments, except for upland rainfed rice in Brazil. Despite growing in flooded soil during much of the growing season, lowland rainfed rice is also exposed to water deficit and/or excess flooding during a significant portion of the cropping season28, leading to lower and less stable yield potential compared with irrigated rice (Supplementary Figs. 3A and 4). Similar yield trends relative to water regime and rice cropping intensity are observed for comparisons based on average farmer yields (Supplementary Fig. 3B).
Expressing average actual farmer yields as a percentage of yield potential estimates the magnitude of yield gap closure, which in turn offers an objective measure of the degree to which rice growers efficiently utilize solar energy, water, and nutrient resources. At a global scale, average rice yield represents 54% of yield potential, with a wide range of yield gaps across rice systems (Fig. 1 and Supplementary Fig. 5). For example, irrigated rice systems in northern China, Australia, and California have reached ca. 75% of the yield potential. At the other end of the spectrum, average yields are low for lowland rainfed rice in Sub-Saharan Africa and upland rainfed rice in northern Brazil and represent 20–40% of the yield potential. About two thirds of global rice harvested area have yields lower than 75% of yield potential; the latter is considered a reasonable yield gap closure target for farmers20. Overall, our analysis indicates substantial room to increase global rice production on existing planted area via improved agronomic management.
Benchmarking resource-use efficiencies. We looked at key environmental and resource use metrics associated with rice cropping systems, including global warming potential (GWP), water supply (sum of irrigation plus in-season precipitation), pesticide use, nitrogen (N) balance, and labor inputs. Despite a strong positive correlation between the degree of yield gap closure and total input use per unit area expressed as GWP, high-yield systems have lower GWP on a yield-scaled basis due to higher resource-use efficiency (Fig. 2A, B). For example, high-yield systems in northern China, Australia, and California have larger GWP per hectare, but smaller yield-scaled GWP than low-yield, low-input systems in Sub-Saharan Africa. An implication from this finding is that, to reach a given grain production target, low-input systems would need larger crop production area, which, ultimately, would lead to a larger environmental impact compared to high-input, high-yield systems. These results are consistent regardless of whether GWP is considered on a per-crop or annual basis (Supplementary Fig. 6).
Across the 32 cropping systems, major contributors to GWP are CH4 emissions from rice growing in lowland systems with soils kept purposely flooded (51%), emissions associated with manufacturing, packaging, and transportation of agricultural inputs (30%), and soil N2O emissions derived from N application (19%) (Supplementary Fig. 7). Variation in CH4 emissions across cropping systems are mostly associated with differences in water and straw management and length of the cropping season cycle, from field preparation to harvest. In the case of upland rice production in Brazil, rice is grown in aerobic (non-flooded) soil conditions, which reduces CH4 emissions and GWP (Fig. 2A, B). In contrast, major drivers for differences in CH4 emissions across flooded-rice systems are length of the rice crop growing cycle and straw management (Supplementary Fig. 1 and Supplementary Table 3). Cropping systems where straw is left in the field and/or with long crop cycle length (e.g., Australia) have higher CH4 emissions and GWP, on a per-crop basis, than systems where crop residues are removed from the field and/or with shorter duration of the rice crop growing cycle (e.g., Indonesia). The positive effect of shorter crop cycle length at reducing CH4 emissions is not apparent on an annual basis because short crop cycle length is associated with tropical rice systems, which, in turn, have a higher number of rice crops per year.
There is no relationship between the degree of yield gap closure and water supply (p = 0.50), probably because water supply is sufficient to meet crop water requirement in most cropping systems (Fig. 2C and Supplementary Fig. 6; Supplementary Table 4). For a similar degree of yield gap closure, there is a large range in water supply due to differences in climate among cropping systems28. For example, water supply is ca. 2x larger in the semiarid climate of California, USA compared with the humid southern USA. Similarly, there is large variation in yield gap at any water supply, with rainfed rice exhibiting a larger gap compared with irrigated rice. The yield-scaled water supply follows a relationship with the degree of yield gap closure similar to that for yield-scaled GWP (r=-0.75; p < 0.01), with smallest values corresponding to cropping systems with small yield gaps (Fig. 2D). Many of these systems are located in semiarid environments (e.g., California, Egypt, and Australia) where crops are likely to be fully irrigated, with little precipitation to supplement crop water demand, and with high yield potential due to high solar radiation (Fig. 2D and Supplementary Fig. 4). Given the low production risk and favorable conditions, these systems are also likely to have a smaller yield gap. Assessing the long-term sustainability of irrigated rice systems in water-scarce environments would benefit from expanding the analysis to larger spatial scales (e.g., watershed) and accounting for recharge rates and stream flows. While incomplete, our study makes a first step on this direction by benchmarking the efficiency in using water resources to produce rice at field scale.
In the case of pesticides, there was a positive relationship between the degree of yield gap closure and the number of applications (r = 0.52; p < 0.01) (Fig. 2E). It is difficult to interpret this relationship considering likely differences in edaphic and climatic environments and the severity of biotic stresses. Higher pesticide use in cropping systems with small yield gap was possibly related to greater pest and disease pressures as a result of large and denser leaf canopies that are achieved with improved plant nutrition30. Likewise, systems with high cropping intensity (i.e., double and triple rice) in tropical areas receive a larger number of insecticide and fungicide applications per crop (up to nine) as in the case of Indonesia and Vietnam (Fig. 2E). There are also labor cost considerations. In contrast to tropical systems, where rural labor wage is low and weeds are mainly removed manually, chemical control prevails in non-tropical cropping systems (Supplementary Table 3). Due to these interrelationships, the relationship between yield-scaled pesticide applications and yield gap closure was not as clear as for GWP and water supply (Fig. 2F), although a similar trend was apparent when cropping systems from Sub-Sahara Africa were excluded from the analysis (r=-0.44; p < 0.05).
Relationships between yield, N input, and N balance (the latter calculated as N input from fertilizer, manure, and fixation minus N removal) are of interest because N is typically the most limiting factor in rice cropping systems and also an important source of environmental pollution31,32. In general, a large positive N balance is a strong indicator of inefficient fertilizer use and potential reactive N losses into the environment, while a negative N balance suggests high risk of soil N mining that degrades soil quality33. For example, data from cereal systems show that potential N losses increase substantially when N balance exceeds 75 kg N per ha33–35. Our analysis shows a positive linear relationship between the degree of yield gap closure and N input (r = 0.73; p < 0.01) and, to a lesser degree, with N balance (r = 0.43; p < 0.05) (Fig. 3A, B). Cropping systems with smallest yield gap tend to have N inputs and N balance greater than 150 and 50 kg N ha− 1, respectively, with a yield-scaled N balance ranging between zero and 20 kg N Mg− 1 grain (Fig. 3). Within this group of cropping systems with small yield gaps, some have a relatively small N balance (50–75 kg N ha− 1) and yield-scaled N balance (< 10 kg N Mg− 1 grain) as in California and Australia (Fig. 3B, C). In contrast, other cropping systems with small yield gaps, such as southern USA and southern and central China, exhibit a relatively large N balance (> 100 kg N ha− 1) and yield-scaled N balance (> 15 kg N Mg− 1 grain), suggesting room for reducing N input and N balance without yield penalty.
The relationship between average yield and yield-scaled N balance follows a curvilinear pattern (r = 0.64; p < 0.05), with larger yield gaps at both ends of the range of yield-scaled N balance (Fig. 3C). On the one hand, there are a number of cropping systems in Sub-Saharan Africa and South-East Asia exhibiting negative N balance, suggesting soil N mining over time (Fig. 3 and Supplementary Fig. 8). These systems would clearly benefit from a larger N input or other methods to improve soil N supply. On the other hand, there is a group of systems with N input > 150 N ha− 1 and large yield gaps, resulting in a large positive N balance on both area and yield-scaled basis, which is the case of several cropping systems in South and South-East Asia. In these systems, it seems feasible to reduce N inputs while maintaining yields or, perhaps more interestingly from a crop production perspective, to increase current yield with the same N input, in both cases leading to lower environmental impact and greater profit. This global analysis also shows that, while a small yield-scaled metric is desirable in the case of GWP, water, or pesticides, it is preferable that the (yield-scaled) N balance is maintained within an acceptable range (i.e., not excessively high or excessively low) to avoid both soil N mining and reactive N losses. This range seems to correspond to a yield-scaled N balance between 5 and 10 kg N Mg− 1 grain (Fig. 3C).
Labor in rice cropping systems. Labor use varied more than 100 times (ranging from 7 to 900 h ha− 1 crop− 1) across rice cropping systems, with degree of mechanization explaining differences among countries (Fig. 4A and Supplementary Table 3). Although it is difficult to separate cause-effect, the analysis suggests that large field size, high mechanization level, and direct seeding are intrinsically linked. For example, labor input is less than 40 h ha− 1 crop− 1 in highly mechanized systems in the USA, Australia, and Uruguay, where field size is greater than 40 ha and most rice is direct seeded (Fig. 4A and Supplementary Table 3). In contrast, labor input is more than 400 h ha− 1 crop− 1 (and up to 900 h ha− 1 crop− 1) in less mechanized systems such as those in Sub-Saharan Africa and Asia, where field size is smaller than 3 ha and most of the rice is transplanted.
One can still find large differences in yield-scaled labor (i.e., number of hours per unit yield) for a given labor input and there is a negative association between degree of yield gap closure and yield-scaled labor (r=-0.71; p < 0.01), which is consistent for both less mechanized and highly mechanized systems (Fig. 4B). For example, in the case of less mechanized systems, yield-scaled labor is smaller in South-East Asia and China (average: 110 h Mg− 1) compared to Sub-Saharan Africa (> 200 h Mg− 1). Similar variation is observed across highly mechanized systems, with low yield-scaled labor in the USA and Australia compared to South America (1 versus 12 h Mg− 1). To summarize, our study shows no trade-offs between yield gap closure and labor requirements while yield-scaled labor decreased with smaller yield gaps in both labor intensive and highly mechanized cropping systems. This finding suggests that a simultaneous improvement in yield and labor productivity is possible, which is relevant in the context of increasing labor wages and shrinking rural population in developing countries36,37.
Overall system performance. We computed an overall performance index for the 32 rice cropping systems in our study (Fig. 5). Our analysis shows that the overall system performance is better in non-tropical versus tropical regions, probably due to inherent differences in soil and climate endowments leading to different resource-use efficiency and input requirements (e.g., higher nutrient and pesticide requirement per unit of yield in tropical environments) 38. Still, one can identify systems that outperform other systems within each environment, as it is the case of California, Australia, and northern China (non-tropical regions), and Vietnam and Thailand (tropical regions). The analysis shown in Fig. 5 is also useful to identify, for a given country, where largest opportunities exist (yield gap, resource-use efficiency, or labor) to improve the overall performance of the cropping system. For example, pesticide use and N balance per unit of production is disproportionally higher in a number of cropping systems in South-East Asia and South Asia.