Worldwide increases in nitrogen (N) inputs to croplands have been and will continue to be an important contributor to growing more food. But a substantial portion of N inputs to croplands are not captured in harvested products and leave the field, contributing to air and water pollution. Whether the proportion of N inputs captured in harvest grows, shrinks, or remains unchanged will have important impacts on both food production and N pollution. We created a new global N input database (fertilizer, manure, fixation, deposition, and residues) that enables evaluation of trends in nitrogen use and recovery by country and by crop from the 1960s through 2007. These data show that despite growth in yields and increased N fertilization, differences in efficiency of N use between Organisation for Economic Co-operation and Development (OECD; http://www.oecd.org) and other countries have persisted over nearly 50 years and exhibit no sign of convergence. The high yield, high nitrogen input systems characteristic of rich countries have released large amounts of reactive N to the environment but have operated with greater efficiency—recovering a greater portion of added N in crops. Aggregate yields in OECD countries are 70% greater than in non-OECD countries on N input rates just 54% greater. Variation in recovery efficiency between countries suggests that there is scope for improvements through enhanced N delivery and capture in the world's low-yielding croplands and that increasing efficiency of N use is an important component of meeting food demand in the future.
 Growing N inputs are expected to continue to contribute significantly to increased crop production [Tilman et al., 2002], particularly in developing countries [Organisation for Economic Co-operation and Development-Food and Agricultural Organization of the United Nations, 2009]. Increasing N inputs often offers immediate remedy to low crop yields [Liu et al., 2010], but field experiments consistently demonstrate that the proportion of added N that is recovered in harvested products (recovery efficiency – (REN)) tends to decline with increasing N inputs [Cassman, 1999]. High N input systems, sometimes characterized as leaky with respect to N [Drinkwater and Snapp, 2007], may lead to diminishing food production returns as fertilizer N inputs grow [Cassman et al., 2003; Tilman et al., 2002] and accelerating release of reactive N that pollutes surface and groundwater affects sensitive ecosystems and increases atmospheric N2O (an important greenhouse gas) [Smil, 1999b].
 But unlike in field-scale N input experiments, agronomic advances capable of ameliorating constraints other than N can lead to increasingly efficient recovery of N inputs in harvested products [de Wit, 1992]. As N inputs change over years and decades, other aspects of crop production technology do not remain static—harvest index and other crop characteristics, availability of micronutrients, disease and pest control, other inputs like micronutrients and irrigation, and harvest efficiency have all evolved. Fertilizer is a valuable resource—its manufacture alone requires about 2% of the world's annual energy supply [Smil, 2000]—and economic incentives to reduce input costs could lead to spontaneous, ongoing increases in REN [Frink et al., 1999; Waggoner and Ausubel, 2002]. Because release of reactive N grows with increasing N inputs, total release of reactive N to the environment is likely to continue to grow [Galloway et al., 2008]. But continued increase in the efficiency of N use in croplands could enhance food production while reducing the negative impacts that arise from continued growth in N inputs to food production systems [Galloway et al., 2008; Spiertz, 2010].
 Feeding a growing population that is increasingly affluent is likely to continue to drive increases in total N inputs to crop production. The extent to which N pollution accelerates with increasing yields and whether release of reactive N can be mitigated to some degree by increasing efficiency of N use has important implications for future N pollution and food production. The purpose of this work was to evaluate past trends in global N use to understand past trends in how efficiently N has been used in the world's croplands, how those trends vary from country to country, and whether the recent past indicates either widespread diminishing yield returns per unit N released to the environment or increases in efficiency of N use over time. Because no global data set exists that allows us to address these issues, we synthesized data from a variety of sources to create a new database with information on total N inputs (fertilizer, manure, and fixation, as well as deposition, and residues which were not used for our REN calculations) for the six main cereal crops (barley, maize, millet, rice, sorghum, and wheat) plus soybeans and pulses, by country over the past 50 years. We investigated changes in the types and amounts of N inputs and how effectively N inputs have been used by the rich Organisation for Economic Co-operation and Development (OECD) countries, the large emerging economies (the BRICS—Brazil, Russia, India, China, and South Africa), and the developing countries in different world regions. Our characterization of past trends and patterns provides a robust foundation for forecasting future release of reactive N to the environment.
 We estimated national-level, crop-specific N fertilizer input rates through time by scaling information on crop-specific rates (N inputs per unit area) from multiple independent sources [Food and Agricultural Organization of the United Nations (FAO), 2006, 2010; Heffer, 2009], divided by the average N application rates for the country in the specified year [FAO, 2009]. We estimated the scaling ratio using Bayesian inversion [Clark, 2005; Ogle and Barber, 2008] constrained by national N fertilizer consumption data [FAO, 2009]. Thus, our analysis integrated bottom-up observations (crop- and country-specific surveys of producer practices) with top-down constraints (country-level N inputs). The probability of each crop-specific value was based on the sum of all crop-specific estimates within each country and constrained by multiple observations of crop-specific fertilizer application rates. Total N fixation was estimated using crop-specific fixation rates [Herridge et al., 2008] and areas [FAO, 2009]. Total manure N application to crops was calculated based on livestock numbers [FAO, 2009], average manure production per head by livestock type [Bouwman et al., 1997], proportion of manure available for use on croplands by animal type and by region [Mosier et al., 1998], N lost through NH3 volatilization [Bouwman et al., 1997], and proportion of manure added to crops [Smil, 1999b]. Harvested N was calculated from data on grain production [FAO, 2009] and N concentrations by crop type.
2.1 Fertilizer N Inputs
 We estimated fertilizer N inputs by compiling information on national fertilizer consumption data published by the UN Food and Agricultural Organization [FAO, 2009] and fertilizer N application rates by crop by country [FAO, 2004, 2010; Heffer, 2009]. The latter included 2337 independent estimates of fertilizer application rates (spanning years 1986 to 2010) from 98 countries that consume 99% of global N fertilizer (Table S1 in the supporting information), covering 65% of country and crop combinations. Our estimates of fertilization rates for each country (i) and crop (c) combination (frli,c) were constrained by the model estimate of total fertilizer consumption for the country (TNli) and cropland area for the country and crop (Ai,c), where TNli = Σfrli,c × Ai,c. All fertilization rates were standardized to a common year (1998) to perform calculations. We estimated country- and crop-specific fertilization rates with Bayesian Markov Chain Monte Carlo (MCMC) modeling. The fertilizer application rate observations were used as prior information to inform the posterior estimates for those combinations. For the combinations where no information was available, we sampled points from the global distribution of values for the specific crop (c). For the crops on which we had no information for any countries (totaling 9% of global crop area), we created a “remainder crop” category and estimated an average N application rate (sampled from the global distribution of values across all crops). The conditional distribution of the MCMC model was as follows:
where fri,c is the fertilizer application rate for the country (i) and crop (c) combination, TNi is the total N consumption for country (i), and the latent variables (l) are the “true” value predicted by the model given the estimated uncertainties (σ). The uncertainty estimates were constrained by an uninformative uniform prior greater than zero.
2.2 Nitrogen Inputs From Manure, Deposition, and Fixation
 Animal manure N inputs were calculated by estimating manure production and accounting for the fraction added to croplands. Production was estimated by multiplying livestock numbers for each country and year [FAO, 2009] by average manure N production per head per year for each livestock type [Bouwman et al., 2005]. Manure N inputs into global croplands vary by region and by livestock type based on manure management practices. Some manure is deposited directly onto pasture and range. Following previous analyses [Smil, 1999b], we assume that this manure is not collected and used for crop application. Of the manure that is deposited in confinement, some is used for fuel, making it unavailable for crop application. The remaining manure was assumed to be available for application. We estimated animal manure N inputs into croplands by multiplying manure N production by a manure availability factor for each livestock type and world region (Table 1) [Mosier et al., 1998] and accounting for NH3 volatilization during storage [Bouwman et al., 2005]. We assumed that 90% of the manure that was available for application was applied to croplands [Smil, 1999b]. We calculated N deposition inputs with regional estimates of deposition rates onto croplands [Dentener et al., 2006] multiplied by total crop area [FAO, 2009]. We calculated N fixation inputs as the sum of legume-bacterial associations and free-living soil bacterial fixation. Legume fixation was calculated with legume N content above and belowground [FAO, 2009; Herridge et al., 2008] multiplied by the percent of N derived from atmosphere by crop [Herridge et al., 2008]. Soil fixation inputs were estimated with global average fixation rates per area, accounting for higher rates in paddy rice [Smil, 1999b]. Fixation inputs were allocated to crops based on crop identity (i.e., N fixers versus non-N fixers and rice) and area. Animal manure and atmospheric deposition inputs were allocated to crops equally by area.
Table 1. Fraction of Animal Manure Production Available for Cropland Applicationa
 Crop residue N was allocated based on region- and crop-specific recycling factors. We estimated crop residue production by multiplying crop production data [FAO, 2009] by a residue conversion factor for each crop [Lal, 2005; Smil, 1999a]. We assumed that there were regional differences in management and estimated a residue recycling factor for each world region. Estimates were based on harvested yield multiplied by crop-specific harvest index factors [Smil, 1999a] and reduced according to regional estimates of the proportion of crop residues available for livestock consumption. Regional proportions were calculated from multiple sources (Table S2 in the supporting information) with either direct information on the fraction of residues used for livestock or by subtracting the fraction of other uses of residues, including biomass energy, building materials, and fertilizer recycling. These factors account for residue burning removals and other noncropland human uses. Crop residue N inputs were estimated by multiplying the estimate of residue production by residue N concentrations for each crop [FAO, 2003; Lal, 1995; Smil, 1999a] and the recycling factor for each world region.
2.4 N Recovery Efficiencies
 Annual N inputs from fertilizer, manure, and fixation were summed to estimate total external N inputs specific to each country, year, and crop combination. Nitrogen in harvested products was calculated as for N in residues by multiplying yields [FAO, 2009] by N concentrations for the harvestable component of each crop [FAO, 2003; Lal, 1995; Smil, 1999a]. We calculated N recovery efficiency (REN) as N in harvested grain as a proportion of external N inputs (fertilizer, manure, and fixed N). Recovery efficiency is one of many means of calculating N use efficiency, each of which has benefits and detractions [Ladha et al., 2005; Semenov et al., 2007]. Our estimate of REN does not account for variability in the supply of N from soil N mineralization, but REN does allow comparison of the benefits (harvested N) versus costs (N inputs), it enables evaluation based on the N inputs that are managed, and it can be calculated using widely available data [Mosier et al., 2004; Robertson and Vitousek, 2009].
3.1 Trends in Yields and N Inputs
 Between 1961 and 2007, worldwide N inputs increased (+134%) faster than crop yields (+120%) (Figures 1a, 1b, 2a, 2b, 3a, and 3b). This trend was observed in most regions (Figure 4) and countries, with both increasing across all regions and for most crops grown in most countries (90 and 81% of cases for N inputs and yields, respectively). Yield growth rates exceeded N input growth rate for four regions (OECD Europe, non-OECD Europe, Asia, and Oceania), India, and the former Soviet Union. Growth in N inputs resulted in the lowest proportional increases in yield in the Middle East (both OECD and non-OECD), OECD Oceania, the non-OECD Americas, and China (Figure 4). At the national level, changes in yield and N inputs usually moved together—either both increasing (575 of 743 crop-country combinations) or both decreasing (23 crop-country cases). Yield increases were observed in 41 crop-country cases in which N inputs decreased, most of which were in Africa (32 crop-country cases from 14 countries) and Asia (13 crop-country cases from 10 countries). Only seven crop-country cases of increasing yields with declining N inputs were from OECD countries (former Czechoslovakia, Denmark, Germany, and Norway). There were 104 crop-country cases in which N inputs increased but yields decreased, despite an average increase in N inputs of 177% in those countries. The vast majority of those cases (98/104) were in non-OECD/non-BRICS countries, with most of those in Africa (56/104) and the Americas (20/104).
 Wheat crop yields exhibited the greatest proportional increase over time (+143%) followed by maize (+140%), rice (+103%), and soybeans (+100%). Increased yields were strongly related (r2 = 0.887; P < 0.05) to increased N inputs. Both yield and N input growth were greater for the BRICS countries (+164% and +179% for yield and N inputs, respectively) than for OECD (+96% and +115%) and other countries (+93% and +96%). Yields averaged across OECD countries were consistently greater than averages for the non-BRICS/non-OECD countries. Growth rates for yield and for N inputs were both slower for pulses, millet, barley, and sorghum than for wheat, maize, rice, and soybeans.
 Yields and N inputs varied substantially across regions within each of the three country groupings. Among the OECD countries, yields and N inputs for countries in Asia and Europe tended to be greater than those in the Middle East and Oceania, with countries from the Americas in between. Grain yields among the BRICS countries were similar at the beginning of the time series, but yield growth has been substantially greater in China and has been accompanied by the fastest growth in N inputs among the BRICS. Yields in the former Soviet Union (Figure 2a) doubled between 1961 and 2007, but much of the growth was realized by the mid-1980s, and a decline in the N input rate since then has been accompanied by slow growth rates for yields (Figure 2b). A similar pattern was exhibited in the Eastern European countries, with even more limited growth in yields after the early 1980s.
 The composition of N inputs changed over time, with the proportion of external N inputs from fertilizer doubling while the contributions of fixation (−27%) and manure (−57%) both declined substantially. Fertilizer N as a proportion of total N grew the most for the BRICS countries (from 14% to 63% of external N inputs). In the early 1960s, fertilizer contributed one third or more of external N inputs in 75% of OECD countries but just 29% of worldwide N inputs (Figure 5). In 2006, mineral fertilizer comprised 63% of total N inputs and contributed the majority of external N in more than 60% of all countries. Fertilizer contribution to total external N inputs increased in most (87%) countries, while the contributions of fixation and manure N declined in most countries (74 and 72%, respectively). Manure N contributions to total external N inputs declined by 57% worldwide. The only region in which manure increased as a proportion of total N inputs was in OECD Asia (Japan + Korea, both of which exhibited increases). Biological N fixation contribution to total external N declined within the BRICS and non-OECD/non-BRICS groups but increased within the OECD countries—reflecting substantial growth in soybean area (which contributed 45% of growth in crop area) and slower growth in fertilizer N inputs than in other regions.
3.2 Nitrogen Recovery Efficiencies
 Globally, about 40% of N inputs to crops are recovered in harvested products, and the proportion of N recovered in grain (REN) was about the same at the beginning of the time series as it was at the end. REN averaged across OECD countries was consistently greater than the average in the non-OECD countries, which was about the same as the BRICS-wide average (though there was substantial variation in REN across the BRICS countries). Trends over time and current REN values varied within each country group. For example, across the BRICS, Brazil, India, and the former Soviet Union saw substantial increases in REN, particularly the former Soviet Union which went from 77% recovery to 125% (Figure 2c). Over the same period, South Africa had a slight decline in REN, while REN in China declined from 37% to 29%.Geographic regions within both the OECD and non-OECD/non-BRICS groups also showed differences in trends with the Middle East (OECD and non-OECD) and OECD-Oceania exhibiting the largest declines and European countries (OECD and non-OECD) exhibiting the greatest increases (Figures 1c and 3c).
 The top 10 countries produced 70% of total grains and legumes harvested in 2005 (and 68% in 1963); all exhibited growth in N fertilization rates, while between the 1960s and 2000s, REN increased for six (USA, the former Soviet Union, Brazil, Argentina, Indonesia, France, and Canada) and decreased for the others (China, India, Indonesia, and Bangladesh) (Figure 6). For many of the large producers, increases in REN have occurred despite continued increase in N inputs. In contrast, increased REN in Argentina and the former Soviet Union were associated with recent stagnation or declines in fertilizer N input rates.
4.1 Pathways of REN and Agricultural Development Through Time
 Despite large increases in yields, growth in N inputs, and substantial shifts in type of N inputs—away from manure and biologically fixed N and toward mineral fertilizer N—there was little net change in REN between the early 1960s and 2007 within all three country groupings. Gaps in N recovery efficiency between the OECD countries and the non-OECD/non-BRICS countries that were evident in the early 1960s have persisted and show no signs of convergence. Neither the global REN record nor those of the three regional groupings exhibit a clear tendency toward accelerating N pollution costs of increasing yields or synchronized increases in yield and N recovery efficiency. But the record of country- and regional-level REN trends over the last 50 years does contain clear examples of both as well as examples of trends that follow two other distinct patterns: (1) reversion to high REN indicative of inadequate N inputs (exemplified by many of the non-OECD Eastern European countries and the former USSR) and (2) low REN values that have changed little over time.
 Nitrogen recovery efficiency declined over time for several countries and regions. These were accompanied in most cases by increases in N inputs and in yields. Declining REN could be driven by a shift from dependence on mining soil mineral N to an increased reliance upon external N inputs [Vitousek et al., 2009]. This seems likely to be the case in several countries and regions in which N inputs and yields are large, and REN declined from values in the early 1960s that were greater than 1 (i.e., in which N harvested exceeded N inputs, e.g., Argentina, Australia, Canada, OECD-Middle East). Convergence with regional mean REN values and previous spatial analysis of patterns of soil nutrient imbalances [Liu et al., 2010; MacDonald et al., 2010] are consistent with this explanation. Declining REN could also be driven by excessive external N inputs. This seems to be the case in some countries, in particular for China, where external N inputs are currently large, where N inputs have grown rapidly, where rice contributes substantially to grain production (our data show that within-country REN for rice was low compared to the other staple crops—maize and wheat), and where crop production seems capable of being increased substantially without further growth in N inputs [Chen et al., 2011]. Substitution of external N inputs for soil mineral N may lead to more sustainable agricultural production, but in cases where low REN is a sign of excessive N inputs, continued N growth will accelerate N pollution [Tilman et al., 2002; Vitousek et al., 2009].
 Globally, there was a tendency for REN to increase from the 1980s to the 1990s and from the 1990s to the 2000s, particularly within the OECD countries. Among the OECD countries, this trend was most pronounced in Europe. Several European countries (e.g., France, Germany, Greece, Italy, Spain) exhibited ongoing increases in REN after reaching nadirs in the 1970s or 1980s. Nitrogen recovery efficiencies tended to be high in these countries initially and increased over time. Both are consistent with previous findings for other components of European agriculture between 1970 and 2000 [Bouwman et al., 2009], though our results suggest that yield returns per unit N input have grown as well. Several other countries, including the United States, followed the same trend which is consistent with a desire in high-input countries to maximize factor productivity [Ausubel and Waggoner, 2008]. A similar pattern was observed for the former Soviet Union and several non-OECD Eastern European countries. Those regions experienced a large rapid increase in REN, but unlike the countries in OECD Europe, increasing REN was accompanied by declining N inputs as those countries underwent political and economic transition. Within both regions (and most countries therein), yield growth continued at a slower rate, so increases in REN signify reversion to mining of soil N, consistent with other studies illustrating negative nutrient balances in these regions [Liu et al., 2010; MacDonald et al., 2010].
4.2 Uncertainties and Limitations of Our Approach
 Our approach to developing the global N input database is similar to those conducted previously, the main exception being our integration of more data sources and using a Bayesian statistical approach for synthesis which enabled integration of both bottom-up data and top-down constraints. Our model allowed us to estimate fertilizer allocation in places where no crop-specific information was available based on our knowledge of allocation practices in other places. The separation of observation errors and the Bayesian hierarchy enabled us to make statistically rigorous estimates in places where little information is available. Without a structured model like ours, it is not possible to share information across countries, regions, and crops while allowing for uncertainty in all of the reported values. Our estimates of N inputs are comparable with estimates from other sources. Our global estimates of fertilizer N applied to croplands (29.1, 79.1, and 82.5 Tg N yr−1 in 1970, 1995, and 2000, respectively) are consistent with estimates by Bouwman et al. (29 and 83 Tg N yr−1 in 1970 and 2000), Smil [1999b](78 Tg N yr−1 in the mid-1990s), and Potter et al. (70.2 Tg N yr−1 for 90% of crops in 2000) and similar to smoothed estimates of world fertilizer consumption from the UN Food and Agricultural Organization (31.3, 77.6, and 82.7 Tg N yr−1 in 1970, 1995, and 2000, respectively)[FAO, 2009] but somewhat larger than those of Liu et al. (67.8 Tg N in 2000). Comparison with manure N inputs to cropland is more challenging because estimating manure N available for use on croplands requires estimation of manure production, proportion produced in confinement, proportion used for other purposes, and volatile N losses. Given similar approaches, consistency between our global manure N input estimates and those of Smil [1999b] (25.0 versus 18 Tg N yr−1 in the mid-1990s) and Liu et al. (23.9 versus 17.3 Tg N in 2000) is not surprising. Our estimates of manure production are generally consistent with previously published estimates [see Potter et al., 2010, Table 2] as are our estimates of manure N available for croplands [Bouwman et al., 2009; Sheldrick et al., 2003].
 A growing body of information on contemporary parameters required to assess N production exist, but information about changes over time is limited. In several cases, this required the assumption that these values have not changed over time. However, we know that improvements in crop selection, inoculation, and other factors have altered N fixation rates per unit yield and per unit area, for example [Chen et al., 2013; Ciampitti and Vyn, 2012]. But our ability to extrapolate these changes back in time and across countries is very limited. Data limitations on the availability of manure, residue management, and leguminous N fixation rates could affect the accuracy of our estimates of temporal trends. Our global-scale analysis also precludes analyses of drivers other than the composition of N inputs and crop composition. Several aspects of agronomic management determine how efficiently N inputs are used [Cassman et al., 2002; Cherry et al., 2008; Ciampitti and Vyn, 2012], but such information is unavailable at the same scale and resolution as our N input and N harvest data. Our results are useful, however, in identifying region/crop combinations in which REN is high or has increased over time, potentially directing further investigation into the reasons why. For example, Africa as a whole has remained quite nutrient poor; increased N inputs would surely lead to increased yields. But, despite low N inputs, REN is close to the non-OECD/non-BRICS average. This suggests that investments in technology to enhance REN could also drive increased yields. Increased N inputs and increased REN is clearly the best approach.
4.3 Implications and Conclusions
 Our analysis suggests that many regions and countries within the non-OECD/non-BRICS group simultaneously suffer from nutrient deficiencies while using those sparse N inputs relatively inefficiently. Both yields and N inputs grew slowly over time and tended to be low in these countries, but REN was also low. A greater than average proportion of rice contributes to reduced efficiency overall, but individual crops also exhibited lower REN for these countries than for others. The REN for these countries/crops generally changed little over time. These results show that increased mineral fertilizer is not always the primary driver of low REN and that low input systems are not inherently more efficient. Differences in crop composition and poor condition of cultivated soils in many of these countries likely affect crop responses to N inputs [Sanchez, 2002]. Linked growth of yield and N inputs—and accompanying modest changes in REN over time—suggest that improved agricultural practices, inputs, and technologies—including irrigation water, micronutrients, planting methods, hybrid seeds, better synchronization between N supply and N demand (via fertilizer type, timing, and placement), and improvement in plant genetics that enhance production, yield, ability to capture N inputs, and greater allocation to harvestable products [Cassman, 1999; Cassman et al., 2002; Cherry et al., 2008; Robertson and Vitousek, 2009; Smil, 1999b]—have enabled crops to capture a growing amount of N. However, these practices seem not to have enabled substantial improvements of the efficiency with which N inputs are captured by harvestable products. Though N input rates in non-OECD/non-BRICS countries tend to be low compared with other countries, inefficient use of N inputs leads to relatively more N pollution and limited yield benefits than could be achieved if REN were greater. Even given appropriate technologies to take advantage of greater N inputs, REN could remain low due to lack of market access, high costs of inputs, and misaligned incentives for maximizing agricultural productivity [Palm et al., 2010].
 Because growth in N inputs between 1963 and 2005 was fastest in countries with low REN, the amount of reactive N released to the environment every year has grown at a faster rate than N inputs. Countries with high N input rates release the largest amounts of reactive N to the environment, but if future growth in agricultural production is centered in countries with low REN values that are maintained into the future, global REN would decline, while N2O emissions and other aspects of N pollution would increase per unit food produced [van Beek et al., 2010]. Alternatively, improvements in REN that accompanied yield growth in those countries could achieve more food production and a reduction in environmental impacts. Past trends suggest that even if future growth in N inputs is accompanied by increasingly efficient use of N inputs, reactive N released to the environment will continue to grow.
 Many of the countries with the highest N input rates have consistently harvested a greater fraction of those N inputs in crops. Modest changes in region-wide REN over time suggest that growth of N input rates has been matched by, but not exceeded by, growth in the ability of crops to harvest additional N resources. Our results show that in many regions, agriculture has not reaped the full benefits of increased N inputs. For example, if the non-OECD countries were able to achieve RENs observed for OECD countries, today's crops could be grown while reducing annual release of reactive N to the environment by 10.0 Tg, or N requirements could be reduced by 13.0 Tg yr−1 (16% of total global fertilizer, manure N, and fixed-N inputs to cereal and legume crops). Alternatively, production from the non-OECD countries could be increased by 1904 Tg annually (19%) without increasing N inputs.
 The high yield, high nitrogen input systems characteristic of rich countries release large amounts of N to the environment, but they tend to have operated with greater N efficiency—recovering a greater portion of added N in crops. Aggregate yields in OECD countries are 70% greater than in non-OECD countries on N input rates just 54% greater. Variation in efficiency between and within country groups suggests that there is scope for improvements through enhanced N delivery and capture in the world's low-yielding croplands [Burney et al., 2010] and that increasing efficiency of N use is an important component of meeting food demand in the future [Matson et al., 1997]. Nitrogen recovery efficiency is a simple metric to account the benefits of N in harvested products versus the N input cost of production. Our analysis based on easily observable, widely available data illustrates a wide range of efficiencies and opportunities to mitigate N pollution impacts of food production.
 This work was supported through agreements C/10/049 and C/10/050 with the International Livestock Research Institute, grants from the US Environmental Protection Agency EPA-OAR-CCD-09-07, the National Science Foundation award 0842315, a Queensland Smart Futures Fellowship, and a grant from the Plant Production and Protection Division of the UN Food and Agricultural Organization.