Yield Trends Are Insufficient to Double Global Crop Production by 2050 — TEI-tagged view

Deepak K. RaypersName, Nathaniel D. MuellerpersName, Paul C. WestpersName, Jonathan A. FoleypersName (2013date). “Yield Trends Are Insufficient to Double Global Crop Production by 2050.” PLoS ONEtitle·j.

Object of study — recent spatially resolved yield trends for maizeterm·dom, riceterm·dom, wheatterm·dom, and soybeanterm·dom and their implications for 2050date production

Research question — Are observed crop-yield trends fast enough to double production by 2050date, and where are gains sufficient, slow, or negative?

Contribution — a global, subnational linear-trend analysis and 2050date projection built from roughly 2.5 million crop statistics across about 13,500 political unitsterm·mth

Full text — tagged

Yield Trends Are Insufficient to Double Global Crop Productionterm·dom by 2050date Abstract Several studies have shown that global crop productionterm·dom needs to double by 2050date to meet the projected demandsterm·dom from rising population, diet shifts, and increasing biofuels consumption. Boosting crop yieldsterm·dom to meet these rising demands, rather than clearing more land for agricultureterm·dom has been highlighted as a preferred solution to meet this goal. However, we first need to understand how crop yieldsterm·dom are changing globally, and whether we are on track to double production by 2050date. Using ∼2.5 million agricultural statisticsterm·mth, collected for ∼13,500 political unitsterm·mth across the world, we track four key global crops—maizeterm·dom, riceterm·dom, wheatterm·dom, and soybeanterm·dom—that currently produce nearly two-thirds of global agricultural caloriesterm·dom. We find that yields in these top four crops are increasing at 1.6%, 1.0%, 0.9%, and 1.3% per year, non-compounding rates, respectively, which is less than the 2.4% per year rateterm·dom required to double global production by 2050date. At these rates global production in these crops would increase by ∼67%, ∼42%, ∼38%, and ∼55%, respectively, which is far below what is needed to meet projected demandsterm·dom in 2050date. We present detailed maps to identify where rates must be increased to boost crop production and meet rising demands.

Introduction The world is experiencing rising demands for crop productionterm·dom, stemming from three key forces: increasing human populationterm·dom, meat and dairy consumptionterm·dom from growing affluence, and biofuel consumptionterm·dom [1] – [5] . By 2050date, global agricultural productionterm·dom may need to be increased by 60%–110% to meet these increasing demands [3] , [6] – [7] as well as to provide food securityterm·dom to the ∼870 million now chronically undernourished [8] . The only peer-reviewed estimate [3] suggests that crop demand may increase by 100%–110% between 2005 and 2050date. Numerous authors have suggested that increasing crop yieldsterm·dom, rather than clearing more land for food production, is the most sustainable path for food securityterm·dom [2] , [4] , [9] – [14] . Moreover, crop yieldterm·dom growth has been shown as an effective tool in reducing global poverty and undernourishment, as farmers themselves constitute the vast majority of the poor and the undernourished [15] – [17] .

However, several recent studies indicate that yields may no longer be increasing in different regions of the globe [18] – [23] . Yields are no longer improving on 24–39% of our most important cropland areas [23] . Many of these areas are in top crop producing nations, having rising population, increasing affluence, or some combination of these factors [3] , [5] , [19] , [21] – [23] . This may increase difficulty of meeting future crop production goals but key unknowns remain for developing and targeting strategies: how are crop yieldsterm·dom changing across the world, where gains in crop yieldsterm·dom are able to meet growing demands, and where crop yieldsterm·dom are falling behind.

Here we employ ∼2.5 million statistics from a newly developed crop yieldterm·dom and area harvested database covering ∼13,500 political unitsterm·mth globally from 1961 to 2008, focusing on trends in the recent two decades [4] , [23] . We determine the rates of yield change in each political unitterm·mth for the top four global crops: maizeterm·dom, riceterm·dom, wheatterm·dom, and soybeanterm·dom. These four crops together produce about two-thirds of current harvested global crop calories [3] , [18] . Using these data, we estimate the best-fit linear, non-compounding rates of yield changeterm·mth between 1989 and 2008 for these crops in each of these political unitsterm·mth. Yield change is commonly modeled as a linear function of time [24] – [28] and such models have been used to project future crop yieldsterm·dom [28] – [30] . We provide local, country, and global-scale rates of recent crop yieldterm·dom changes to determine where the rates of yield increase could double production by 2050date, and where they are insufficient. The impact of negative or even slow rates of yield change in these crops could be severe, especially for low-income countries with rapidly rising population. The underlying data, period analyzed, statistical approach, and comparisons of yield projections are described in the Methods section below, with additional details and analysis in Text S1 .

Results The global average rates of yield increase across ∼13,500 political unitsterm·mth are 1.6%, 1.0%, 0.9%, and 1.3% per year for maizeterm·dom, riceterm·dom, wheatterm·dom, and soybeanterm·dom, respectively ( Table 1 , Figure 1 ). A ∼2.4% per year rateterm·dom of yield gainsterm·dom (non-compounding) is needed to double crop production by 2050date. Current ratesterm·dom are thus not achieving this goal. At current ratesterm·dom only ∼67%, ∼42%, ∼38%, and ∼55% increases in maizeterm·dom, riceterm·dom, wheatterm·dom and soybeanterm·dom production, respectively, is possible by 2050date.

Global summary for maizeterm·dom, riceterm·dom, wheatterm·dom, and soybeanterm·dom.

Global projections. Observed area-weighted global yieldterm·mth 1961–2008 shown using closed circles and projections to 2050date using solid lines for maizeterm·dom, riceterm·dom, wheatterm·dom, and soybeanterm·dom. Shading shows the 90% confidence regionterm·mth derived from 99 bootstrapped samplesterm·mth. The dashed line shows the trend of the ∼2.4% yield improvementterm·dom required each year to double production in these crops by 2050date without bringing additional land under cultivation starting in the base year of 2008.

We provide a range of future yield estimates by bootstrap sampling crop yieldterm·dom data at each of the political unitsterm·mth studied for the period 1989 to 2008. The upper bound of the 90% confidence intervalterm·mth ( Table 1 , Figure 1 ) presents a slightly more optimistic scenario, global yields increase at rates of 2.4%, 1.4%, 1.8%, and 2.0% per year for maizeterm·dom, riceterm·dom, wheatterm·dom, and soybeanterm·dom, respectively. Yield trends following this upper bound projection could lead to ∼101%, ∼59%, ∼76%, and ∼84% increased production in these crops, respectively. The lower bound of our confidence intervalterm·mth provides us with a “worst-case scenarioterm·dom,” wherein the global average yield of maizeterm·dom, riceterm·dom, wheatterm·dom, and soybeanterm·dom would increase at 0.8%, 0.5%, 0.1%, and 0.3% per year, respectively ( Table 1 , Figure 1 ). At these rates global production could only increase by ∼34%, ∼21%, ∼4%, and ∼13% for maizeterm·dom, riceterm·dom, wheatterm·dom, and soybeanterm·dom, respectively, by 2050date. Further, the yield trajectory diverges, especially for riceterm·dom and wheatterm·dom from the 2.4% per year rateterm·dom ( Figure 1 ). See Figure S1 for spatial maps of r 2 at each political unitterm·mth and statistical diagnostic tests ( Text S1 and Figures S2 , S3 , S4 , S5 , S6 , S7 , S8 , S9 ).

In the short term, due to population increases from ∼6.7 billion in 2008 to ∼8.0 billion in 2025 [5] , the 1.6% and 1.3% per year global maizeterm·dom and soybeanterm·dom yield improvementsterm·dom may result in no significant change to the per capita global maizeterm·dom and soybeanterm·dom harvests. However, by 2050date there could be an increase. The much lower rates of riceterm·dom and wheatterm·dom yield increases, 1.0% and 0.9% per year, respectively, may result in no change to the per capita riceterm·dom and wheatterm·dom harvests to 2050date. Thus, if we are to boost the production in these top four global crops that are now responsible for directly providing ∼43% of the global dietary energy and ∼40% of its daily protein supply [31] from yield increases alone, we have to immediately determine where and exactly by how much yields are changing. To further understand the yield trend patterns, we also track the rates of yield change within ∼13,500 political unitsterm·mth and report the results at the local and country scales.

Global trends mask the significant variations in the rates of yield changeterm·dom among and within countries ( Figure 2 ). We determine where the within-country yield change rates are ∼2.4% per year or above (i.e. doubling rates), where the rates are lower, and where yields are decreasing. We briefly describe these areas, emphasizing areas with doubling and decreasing rates as these areas define the places with the greatest opportunity to meeting growing demand or where to target investments. See Figure S10 for continuous rate (non-categorical) maps in kilograms/hectare/year/year. The influence of observed yields in 2008 on the percent rate of change is described in Figure S11 and related Text S1 .

Maps of observed rates of percent yield changes per year. Global map of current percentage rates of changes in (a) maizeterm·dom, (b) riceterm·dom, (c) wheatterm·dom, and (d) soybeanterm·dom yields. Red areas show where yields are declining whereas the fluorescent green areas show where rates of yield increase – if sustained – would double production by 2050date.

North and Central America Most of North Dakota and Mississippi, northeastern South Dakota, northwestern Minnesota, and some isolated counties in other United States (U. S.) states are witnessing ∼2.4% per year or greater rates of maizeterm·dom yield gainsterm·dom. Similar doubling rates in maizeterm·dom yields are found in the states of Chihuahua, coastal Sinaloa, most of Michoacán and Guanajuato and isolated areas of few other Mexican states as well as El Salvador. Maizeterm·dom yields are decreasing in parts of the U. S. Great Plains states (southern South Dakota, Kansas, Eastern Colorado and parts of southeastern Texas), eastern Mexico (San Luis Potosí, northern Durango, southeastern Coahuila, Nuevo León and Tamaulipas), and in Haiti and Guatemala ( Figure 2a ). The resultant impact is that the United States has the highest national rates of maizeterm·dom yield improvementterm·dom in this region of the world (in kg/ha/year 2 ) followed by Canada, then Cuba, and Mexico ( Data S1 ). In Central American countries such as Honduras and Nicaragua, where maizeterm·dom now provides ∼27%, and ∼25% of daily dietary energy, respectively [31] , and in Panama (∼7% of dietary energy), the production gains from their slower 0.5% per year yield improvementterm·dom rates could be less than those required to keep pace with their population growthterm·dom (in per capita harvested productionterm·mth terms). In Guatemala, where maizeterm·dom now provides ∼36% of dietary energy [31] the yield trends are already negative (−0.7% per year), and as the population is projected to substantially increase [5] , a steeper fall in the per capita harvested maizeterm·dom could occur.

Riceterm·dom yield doubling rates are found only in some isolated areas of North and Central America. The United States has the highest overall riceterm·dom yield improvementterm·dom rates (1.2% per year), followed by Mexico (1.1% per year). In Nicaragua and Panama where riceterm·dom supplies ∼16% and ∼24% of dietary energy respectively, the per capita riceterm·dom harvests could fall due to their population growthterm·dom [5] outpacing their 0.9% and 0.2% per year riceterm·dom yield improvementterm·dom rates ( Figure 2b ). Elsewhere in the Dominican Republic, Costa Rica, and Haiti, where riceterm·dom provides 16–22% of their daily dietary energy, yields are declining at rates of −0.1% to −0.6% per year. The per capita riceterm·dom production is likely to increase only in Cuba, where riceterm·dom yields are increasing 0.9% per year [31] and the population is projected to fall [5] .

Wheatterm·dom yields are increasing at ∼2.4% per year or greater only in some counties in the U. S., (mainly in eastern South Dakota, parts of Nebraska, northeastern Kansas, western Mississippi, and Louisiana) ( Figure 2c ). Wheatterm·dom yields are decreasing in many parts of the U. S. Great Plains (Montana, western parts of North Dakota, South Dakota, Kansas and Texas and eastern Colorado). In Mexico, areas with doubling rates in wheatterm·dom yields are observed only in the state of Zacatecas. Nationally, wheatterm·dom yields in Canada, United States, and Mexico are increasing at 1.3%, 0.8% and 1.1% per year, respectively.

Most areas in the U. S. show increasing soybeanterm·dom yields, with doubling rates in North Dakota, isolated areas of South Dakota, Nebraska, Mississippi, Louisiana, and Georgia ( Figure 2d ). Soybeanterm·dom yields are decreasing in Kansas, Oklahoma and Texas. Canada and the United States have yields increasing at 0.2% and 1.2% per year, respectively.

South America Most maizeterm·dom areas in South America are achieving doubling rates, with the exception of isolated municípios in Brazil. The overall impact of these varied subnational rates is that maizeterm·dom yields are increasing at 1.7–4% per year in Uruguay, Argentina, Chile, and Brazil and may result in significantly higher per capita maizeterm·dom harvests. Other South American countries such as Venezuela, Peru, Bolivia and Ecuador, where maizeterm·dom provides 2–14% of dietary energy, are achieving yield increases of 1.8–3% per year but due to their population growthterm·dom [5] may result in no significant changes to their per capita maizeterm·dom harvests at least in the short term.

Riceterm·dom is grown throughout in South America and yields are improving in most areas ( Figure 2b ). Riceterm·dom yield rates are at doubling levels however only in the Cesar and Tolima departments of Colombia, and isolated municípios especially in the states of Pará, Maranhão, and Mato Grosso in Brazil and some areas of Paraguay, Uruguay and Argentina. Decreasing rates of riceterm·dom yield are found in northeastern Brazil. The overall impact of these subnational rates of riceterm·dom yield changes is: national riceterm·dom yields are improving fast in Brazil, Peru, Colombia, Uruguay, and Argentina (1.6–2.7% per year). But due to population growthterm·dom [5] , yield increases alone may be unable to boost the per capita riceterm·dom harvested in Peru, Argentina and Ecuador in the short term (∼2025). Elsewhere in Venezuela and Bolivia (where riceterm·dom supplies 8–19% of daily dietary energy and riceterm·dom yields are improving at 1.1–1.4% per year) per capita harvested riceterm·dom could remain unchanged to 2050date. In Suriname, where riceterm·dom provides ∼25% of dietary energy, the very low rates of riceterm·dom yield improvementterm·dom, 0.2% per year, may even lead to decreased per capita riceterm·dom harvests. In Brazil, Uruguay, Guyana, and Paraguay where riceterm·dom now supplies ∼11%, ∼7%, ∼29%, and ∼2% of dietary energy respectively, the per capita riceterm·dom harvested could increase.

Large extents of doubling wheatterm·dom yield rates occur only in Argentina and Chile. Wheatterm·dom yields are decreasing in parts of Peru and in Santiago del Estero in Argentina. The national trends as a consequence: Argentine and Chilean wheatterm·dom yields are increasing at 1.5% and 1.9% per year respectively and may result in increased per capita wheatterm·dom harvests (wheatterm·dom provided ∼25% and ∼30% of dietary energy, respectively). In Brazil, Colombia, Ecuador, and Uruguay, yields are increasing at 0.1–1.5% per year and may lead to unchanged per capita harvests, (8–31% of dietary energy is now supplied from wheatterm·dom in these four countries). Per capita wheatterm·dom harvests could decrease in: Bolivia, Peru, and Paraguay due to lower yield gainsterm·dom of 0.5–1.6% per year.

With the exception of Bolivia and Paraguay, soybeanterm·dom yields are increasing at doubling rates, particularly in many areas of Argentina, Brazil, Venezuela, and Uruguay. Soybeanterm·dom yields are increasing 1.5–2.4% per year in these countries.

Europe Almost everywhere in Europe, except in Moldova, maizeterm·dom yields are increasing but rates of ∼2.4% per year are found only in Portugal, the Czech Republic, and Belarus. In Moldova, southern Romania (counties in south and southwest region) and Pomeranian province in Poland maizeterm·dom yields are deceasing and have led national maizeterm·dom yields to change at −4.9% per year in Moldova, 0.7% per year in Romania, and 1.1% per year in Poland. Due to rising maizeterm·dom yields of 0.8–3.0% per year, the per capita harvested maizeterm·dom could increase in many European countries by 2050date.

Riceterm·dom is harvested in only a few European countries and yields are increasing at 0.2–1.5% per year. See Data S1 for the numbers.

Wheatterm·dom is an important food crop in Europe and harvested in almost all European countries. However, in Eastern Europe (Ukraine, Moldova, southern Romania, Bulgaria, parts of Hungary and Slovak Republic), southern France, and northeastern Spain wheatterm·dom yields are generally decreasing with the exception of a few regions where yield increases are at doubling rates. Consequently, national wheatterm·dom yield improvementsterm·dom in European countries are generally <1% per year, with the exception of Estonia (1.5% per year). Even though wheatterm·dom yield improvementsterm·dom are low, the per capita harvested wheatterm·dom may increase in some of these European countries because of population declines in Estonia, Germany, Latvia, and Lithuania [5] . Unfortunately, in many other European countries, the low production gains from yield improvementterm·dom will likely be offset by increasing population, resulting in nearly unchanged per capita wheatterm·dom harvests. Yields are decreasing in many eastern European countries throughout, where wheatterm·dom comprises 24–36% of the dietary energy.

Soybeanterm·dom yields are increasing at doubling rates only in small areas in Romania, and in central Italy soybeanterm·dom yields are decreasing. See Data S1 for actual national numbers.

Africa Africa is a continent of contrasts with regards to rates of maizeterm·dom yield change. For example, maizeterm·dom yields are increasing ∼2.4% per year in the Nigerian states of Yobe and Adamawa. Similar maizeterm·dom yield improvementterm·dom rates are found in some other isolated areas of WestpersName African nations, Ethiopia, Angola, South Africa, and Madagascar. But maizeterm·dom yields are decreasing in Morocco, Chad, Somalia, Kenya, Zambia, Zimbabwe, Rwanda, Burundi, and Democratic Republic of Congo. Elsewhere, rates of yield improvementterm·dom are lower than population growthterm·dom, suggesting that production per capita is likely to decline. These trends are particularly troubling in countries such as Burundi, Chad, Kenya, Morocco, Rwanda, Democratic Republic of Congo, Somalia, Zambia, and Zimbabwe, where yields are decreasing −0.2% to −7.6% per year, population is rising [5] , and maizeterm·dom accounts for 5–51% of calorie intake. The only African countries that may witness an increase in per capita maizeterm·dom harvests due to faster maizeterm·dom yield increases are Angola, Ivory Coast, and Mozambique, where yields are increasing at rates of 2.9%, 4.1%, and 3.2% per year, respectively.

In Ivory Coast, Togo and Benin in WestpersName Africa, and in Rwanda, riceterm·dom yield changes are at doubling rates. In contrast, yields are decreasing more than 1% per year in Gambia and 3% per year in Nigeria. Nearly 8% of the dietary energy in Nigeria is supplied from riceterm·dom [31] . The per capita riceterm·dom harvests could decrease in almost all the important riceterm·dom consuming African nations, e.g., Guinea, Madagascar, Mali, Nigeria, and Tanzania, unless yields are boosted further. Only in Ivory Coast there could be an increase on account of its ∼2.6% per year yield increases.

Wheatterm·dom, while grown in only an extremely small area of Africa, though in many countries, is generally increasing yields at high rates. In Angola, Eritrea Malawi, Nigeria, Algeria, Sudan, and South Africa, yields are growing at doubling rates (2.4–3.4% per year).

In Nigeria, and Mpumalanga province of South Africa soybeanterm·dom yields are increasing at doubling rates whereas in Zimbabwe, Democratic Republic of Congo, and Rwanda yields are decreasing.

Asia and Australasia Maizeterm·dom yields are generally increasing across Asia. Yield improvementterm·dom rates are currently on track to double production in some parts of Iran, Pakistan, India, China, Indonesia, Bangladesh, Laos, Cambodia, Vietnam, and Turkey ( Figure 2a ). In China, Laos, Philippines, Australia, India, Pakistan, and in Turkey, the per capita maizeterm·dom harvested could remain unchanged in the short term but could increase by 2050date. In a few countries such as Nepal, Kyrgyzstan, Iraq, Afghanistan, Uzbekistan, and in New Zealand the slower rates of maizeterm·dom yield improvementsterm·dom of 1.3%, 0.3%, 0.9%, 1.1%, 1.6%, and 0.5% per year respectively can result in declines in per capita maizeterm·dom harvests on account of population growthterm·dom [5] . The per capita maizeterm·dom harvests could increase in Indonesia, Vietnam, Thailand Iran, Myanmar, Bangladesh, Cambodia, Azerbaijan, and in Bhutan [5] .

Riceterm·dom areas with doubling yield rates are found only in some local areas within Afghanistan, India, Bangladesh, Laos, Vietnam and Cambodia. Significant rates of riceterm·dom yield declines are found in parts of India (especially in parts of Uttar Pradesh, Maharashtra, and Tamil Nadu) and in North Korea. The ∼1% per year overall riceterm·dom yield increase in India could result in no significant change to the overall per capita riceterm·dom harvested but in China this may remain as only a short term (∼2025) problem. Riceterm·dom provides ∼30% and ∼27% of the dietary energy in India and China respectively now. On the other hand in the world's third largest riceterm·dom producer, Indonesia where ∼49% of dietary energy is provided by riceterm·dom, yield improvementterm·dom rates are much lower at 0.4% per year. The other important Asian riceterm·dom producers may behave in the following way: no significant change in per capita riceterm·dom harvests in Pakistan, Nepal, Malaysia, and South Korea; no significant increase in Myanmar, Sri Lanka, Turkey and Bhutan only in the short term, and declines in the Philippines in the long term (∼2050date). However in Afghanistan, Iraq and Australia due to population growthterm·dom outpacing production increases from the 2.4%, 0.4%, and 0.3% rates of yield increase per year there could be declines in per capita harvested riceterm·dom. In North Korea, Kazakhstan, Uzbekistan, and in Turkmenistan yields are declining at −2%, −1.9%, −0.3%, and −1.5% per year respectively. Elsewhere, the per capita riceterm·dom harvested could increase: Bangladesh, Vietnam, Thailand, Cambodia, Laos, and in Iran due to production increases from high rates of yield improvementterm·dom (1–2.6% per year) outpacing their population growthterm·dom [5] , and in Japan due to small yield improvementterm·dom rates (0.5% per year) and small population decreases.

Wheatterm·dom yields are increasing at doubling rates in parts of Iraq, Iran, and Afghanistan, but only in small parts of the top producing countries of China and India. Wheatterm·dom yields are decreasing in many areas of India, especially in the states of Madhya Pradesh and Uttaranchal, and in the countries of Kyrgyzstan, and Mongolia, and in the Beijing province of China. Large areas with wheatterm·dom yield decreases are also found in Queensland, New South Wales, Victoria, and Western Australia, in Australia. The consequence of these wheatterm·dom yield change rates is diverse. Per capita wheatterm·dom production could increase in many countries, including China, Iran, and North Korea, because yield increases exceed projected population increases. In contrast, decreases in per capita harvests could occur in: Afghanistan, Georgia, Iraq, Kyrgyzstan, Mongolia, Saudi Arabia, Yemen, and in Australia.

Soybeanterm·dom yields are increasing at ∼2.4% throughout China, including the provinces of Jilin, Guangxi, and Guangdong, but are decreasing in Yunnan and Ningxia. Yields are increasing at doubling rates in local areas within Maharashtra in India. In contrast, yields are generally decreasing in the neighboring state of Madhya Pradesh. Doubling rates of soybeanterm·dom yields are also found in Laos and Vietnam, but in North Korea, and Cambodia, yields are decreasing.

Discussion and Conclusions Numerous studies have shown that feeding a more populated and more prosperous world will roughly require a doubling of agricultural production by 2050date [1] – [7] , translating to a ∼2.4% rate of crop production growth per year. We find that the top four global crops – maizeterm·dom, riceterm·dom, wheatterm·dom, and soybeanterm·dom – are currently witnessing average yield improvementsterm·dom only between 0.9 to 1.6 percent per year, far slower than the required rates to double their production by 2050date solely from yield gainsterm·dom. This is because yield improvementsterm·dom are below ∼2.4% per year in many areas of our most important agricultural lands. At these rates maizeterm·dom, riceterm·dom, wheatterm·dom and soybeanterm·dom production may increase by ∼67%, ∼42%, ∼38%, and ∼55% respectively, by 2050date globally. There is a 90% chance that the total global production increase from yields alone would be between 34–101% for maizeterm·dom, 21–59% for riceterm·dom, 4–76% for wheatterm·dom, and 13–84% for soybeanterm·dom by ∼2050date. Thus, if these yield change rates do not increase, land clearing possibly would be needed [3] if global food securityterm·dom is to increase or even maintained ( Table 1 ).

We found that the top three riceterm·dom and wheatterm·dom producing nations are witnessing very low yield growth rates. China, India and Indonesia are witnessing riceterm·dom yield increases of only 0.7%, 1.0%, and 0.4% improvement per year. China, India, and the U. S., the top three wheatterm·dom producers similarly were witnessing yield increases of only 1.7%, 1.1%, and 0.8% per year, respectively. At these rates we found that yield driven production growth in India and China could result in nearly unchanged per capita riceterm·dom harvests, but decline steeply in Indonesia.

In many of the smaller crop producing nations, maizeterm·dom, riceterm·dom, or wheatterm·dom yield improvementterm·dom rates are below the 2.4% doubling rate. Unfortunately, a high percentage of total calories consumed in these countries are from these four crops. This is particularly true for maizeterm·dom throughout much of Africa (e.g., Kenya, Zambia, Zimbabwe), Central America (e.g., Guatemala, Nicaragua, Panama), and parts of Asia (e.g., Nepal, Georgia).

Riceterm·dom provides ∼19% of dietary energy globally. Riceterm·dom provides a higher percentage of total calories consumed in countries such as Dominican Republic, Costa Rica, Haiti, Sierra Leone, Nigeria, and North Korea, yet yields are declining, −0.1% to −3.2% per year. Elsewhere riceterm·dom yields are increasing too slowly to overcome the impact of their population growthterm·dom. In some of the world's top riceterm·dom producers, e.g. India and China, the per capita production may remain nearly unchanged. In numerous smaller riceterm·dom producers across the world where riceterm·dom is an important significant provider of daily dietary energy such as in Peru, Ecuador, Bolivia, Benin, Togo, Myanmar, Philippines, Malaysia, South Korea, Nepal, and in Sri Lanka, the per capita production may also remain unchanged.

Wheatterm·dom provides ∼19% of global dietary energy. Wheatterm·dom comprises an even larger portion of the diet in some countries where yields are declining, particularly Eastern European countries of Bulgaria, Hungary, Czech Republic, Moldova, Romania, Slovakia, and Ukraine. In many countries, such as Bolivia, Peru, Paraguay, Afghanistan, and Iraq, wheatterm·dom yield increases are too low to maintain their current per capita harvests.

Our analysis identifies where yield improvementsterm·dom are on track to double production and where investments should be targeted to increase yields. The observed rates of yield change result from several location-specific, socio-economic, and biophysical factors that are described elsewhere [23] . Many studies illustrate that intensification can be unsustainable [32] – [36] , but several notable projects in Africa [37] and elsewhere [38] have shown that sustainable intensificationterm·dom is possible and necessary to boost global crop productionterm·dom.

Clearly, the world faces a looming and growing agricultural crisis. Yields are not improving fast enough to keep up with projected demandsterm·dom in 2050date. However, opportunities do exist to increase productionterm·dom through more efficient use of current arable landsterm·dom [4] and increasing yield growth rates by spreading best management practicesterm·dom and closing yield gapsterm·dom under different management regimes [38] – [42] across the globe. A portion of the production shortfall could also be met by expanding croplandsterm·dom, but at a high environmental cost to biodiversityterm·dom and carbon emissionsterm·dom [4] , [43] – [45] . Alternatively, additional strategies, particularly changing to more plant-based dietsterm·dom and reducing food wasteterm·dom [4] , [46] – [48] can reduce the large expected demand growth in foodterm·dom [3] , [4] .

Methods Data We used annual crop census reportsterm·mth for harvested areasterm·mth and yield from ∼13,500 political unitsterm·mth globally covering 20 years from 1989 to 2008 in this analysis though the database itself covers the years 1961 to 2008. The sum total of these census reports for the 20 years was approximately 1.8 million. Data were collected at three political levels/units depending on data availability: country, state/provinces, and county/district/município/department. Data were not available for all political unitsterm·mth for each year. Details of the number of years data was available and its source is given in the Table S1 . For the political unitsterm·mth where data was missing for some years we estimated crop harvested and yield information using the average of the latest five years of reported dataterm·mth and constraining them with the reported numbers from the higher political unitterm·mth as explained further in Text S1 and previous work [23] .

Population data and its projections per country were from the United Nation's medium variant projectionsterm·mth [5] . Crop production was determined using the projected crop yieldsterm·dom at current observed rates of yield change and harvested areasterm·mth fixed at ∼2007. Per capita harvested productionterm·mth is the ratio of production to populationterm·mth and a greater than ±10% change from ∼2007 is considered as significant either in the short- (2025) or long-term (2050date).

Analysis We linearly regressed 20 years of crop yieldsterm·dom at each of the political unitsterm·mth to determine the average linear rates of yield improvementterm·dom over the observed period. Many previous studies have shown that crop yieldsterm·dom change linearly and have used linear regressionterm·mth to project future crop yieldsterm·dom [24] – [30] . Here we calculate the non-compounding linear percentage rateterm·mth by solving a in Equation 1 ; Y is the yield in the year 2008, 2Y is the yield in 2050date (after 42 years): This gives a rate of 2.38% per year or approximately 2.4% per year. For reported numbers at the local- to country- to global-scale the linear percentage changes are the observed changes using 2008 yields as the base year. The actual changes are provided in Data S1 for each crop and country.

Details of the method used, sensitivity to the number of years analyzed, as well as alternate regression methods are provided in Figures S6 and S9. The advantage of analyzing at high spatial resolutionterm·mth is that yield rates can be summarized for other unique levels. For example, we summarized the results for the Brazilian Legal Amazon ( Figure S12 and Table S2 ). We compared our global numbers with other reported estimates. These comparisons are provided in Table S3 .

Supporting Information Global maps of the coefficient of variation (r 2 ) for maizeterm·dom, riceterm·dom, wheatterm·dom, and soybeanterm·dom when fitted to 20 years of yield information at each political unitterm·mth analyzed.

(TIFF) Click here for additional data file.

Global maps of normality of the data for maizeterm·dom, riceterm·dom, wheatterm·dom, and soybeanterm·dom at each political unitterm·mth analyzed from the Lilliefors test (green colors show where the normality assumptions are not violated at p>0.05 and red colors where they are violated at p≤0.05).

(TIFF) Click here for additional data file.

Global maps of autocorrelation of the data for maizeterm·dom, riceterm·dom, wheatterm·dom, and soybeanterm·dom at each political unitterm·mth analyzed from the Durbin-Watson test (green colors show where the autocorrelation assumptions are violated at p>0.05 and red colors where they hold at p≤0.05).

(TIFF) Click here for additional data file.

Diagnostic plots for a linear fit (r 2 = 0.49, p<0.01) to soybeanterm·dom yield data in the United States. Subplots show a) model fit and standard 95% confidence interval, b) QQ plot, c) residuals versus fitted values, and d) residuals versus lagged residuals. Durbin-Watson test for autocorrelation: p = 0.66. Lilliefors test for normality of yield data: p>0.5.

(TIFF) Click here for additional data file.

Diagnostic plots for a linear fit (r 2 = 0.49, p<0.01) to maizeterm·dom yield data in Angola. Subplots show a) model fit and standard 95% confidence interval, b) QQ plot, c) residuals versus fitted values, and d) residuals versus lagged residuals. Durbin-Watson test for autocorrelation: p<0.05. Lilliefors test for normality of yield data: p>0.5.

(TIFF) Click here for additional data file.

Parsimoniously fitted yields at each of the political unitsterm·mth and using them to project global crop yieldsterm·dom to the year 2025.

(TIFF) Click here for additional data file.

Diagnostic plots for a quadratic fit (r 2 = 0.60, p<0.01) to maizeterm·dom yield data in Angola. Subplots show a) model fit and standard 95% confidence interval, b) QQ plot, c) residuals versus fitted values, and d) residuals versus lagged residuals. Durbin-Watson test for autocorrelation: p = 0.08.

(TIFF) Click here for additional data file.

Consequences of extrapolating linear and quadratic maizeterm·dom yield models for Angola to 2050date.

(TIFF) Click here for additional data file.

Global maizeterm·dom, riceterm·dom, wheatterm·dom, and soybeanterm·dom yield fitted to 15 and 25 years of data and using it to project yields to the year 2025. Dashed lines correspond to analysis using 15 years of data (1994–2008), dotted lines correspond to using 25 years of data (1984–2008), and solid lines correspond to using 20 years of data (1989–2008). Due to the similarity in results in some cases all lines are not clearly distinguishable from each other always.

(TIFF) Click here for additional data file.

Rates of yield change in kg/ha/year/year.

(TIFF) Click here for additional data file.

Year 2008 yields.

(TIFF) Click here for additional data file.

Similar to Figure 1 in the main text but only for the Brazilian Legal Amazon.

(TIFF) Click here for additional data file.

Country data source, number of political unitsterm·mth analyzed per country, time frame and number of official statistics collected per crop for the period 1989 to 2008.

(DOCX) Click here for additional data file.

Current yields, projections and production for the Brazilian Legal Amazon.

(DOCX) Click here for additional data file.

Comparison with the future U. S. crop yieldsterm·dom reported by the USDA-ERS [91] (maizeterm·dom, wheatterm·dom, and soybeanterm·dom were reported in bushels per acre, riceterm·dom in pounds per acres from USDA-ERS and converted to ton/ha), and global wheatterm·dom yields reported by the FAO-OECD [90].

(DOCX) Click here for additional data file.

Additional Data, Model Fitting, Rates, and Inter-comparison information.

(DOC) Click here for additional data file.

Rates observed for each crop and country.

(XLSX) Click here for additional data file.

Acknowledgments We thank George Allez for help in collecting and digitizing a portion of the data, James Gerber for help with Figure 2, and Emily Cassidy for directing us to the Food and Agricultural Organization's Food Balance Sheets. This work also benefitted from the comments of the FoleypersName lab members.

Funding Statement Research support was provided by a grant from the Gordon and Betty Moore Foundation, and by the Institute on the Environment, along with previous funding from National Aeronautics and Space Administration's - NASA's - Interdisciplinary Earth Science program. This work also benefitted from contributions by General Mills, Mosaic, Cargill, Google, PepsiCo, and Kellogg to support stakeholder outreach and public engagement. N.D.M. was supported by a NSF graduate fellowship. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

References Associated Data Supplementary Materials Global maps of the coefficient of variation (r 2 ) for maizeterm·dom, riceterm·dom, wheatterm·dom, and soybeanterm·dom when fitted to 20 years of yield information at each political unitterm·mth analyzed.

(TIFF) Click here for additional data file.

Global maps of normality of the data for maizeterm·dom, riceterm·dom, wheatterm·dom, and soybeanterm·dom at each political unitterm·mth analyzed from the Lilliefors test (green colors show where the normality assumptions are not violated at p>0.05 and red colors where they are violated at p≤0.05).

(TIFF) Click here for additional data file.

Global maps of autocorrelation of the data for maizeterm·dom, riceterm·dom, wheatterm·dom, and soybeanterm·dom at each political unitterm·mth analyzed from the Durbin-Watson test (green colors show where the autocorrelation assumptions are violated at p>0.05 and red colors where they hold at p≤0.05).

(TIFF) Click here for additional data file.

Diagnostic plots for a linear fit (r 2 = 0.49, p<0.01) to soybeanterm·dom yield data in the United States. Subplots show a) model fit and standard 95% confidence interval, b) QQ plot, c) residuals versus fitted values, and d) residuals versus lagged residuals. Durbin-Watson test for autocorrelation: p = 0.66. Lilliefors test for normality of yield data: p>0.5.

(TIFF) Click here for additional data file.

Diagnostic plots for a linear fit (r 2 = 0.49, p<0.01) to maizeterm·dom yield data in Angola. Subplots show a) model fit and standard 95% confidence interval, b) QQ plot, c) residuals versus fitted values, and d) residuals versus lagged residuals. Durbin-Watson test for autocorrelation: p<0.05. Lilliefors test for normality of yield data: p>0.5.

(TIFF) Click here for additional data file.

Parsimoniously fitted yields at each of the political unitsterm·mth and using them to project global crop yieldsterm·dom to the year 2025.

(TIFF) Click here for additional data file.

Diagnostic plots for a quadratic fit (r 2 = 0.60, p<0.01) to maizeterm·dom yield data in Angola. Subplots show a) model fit and standard 95% confidence interval, b) QQ plot, c) residuals versus fitted values, and d) residuals versus lagged residuals. Durbin-Watson test for autocorrelation: p = 0.08.

(TIFF) Click here for additional data file.

Consequences of extrapolating linear and quadratic maizeterm·dom yield models for Angola to 2050date.

(TIFF) Click here for additional data file.

Global maizeterm·dom, riceterm·dom, wheatterm·dom, and soybeanterm·dom yield fitted to 15 and 25 years of data and using it to project yields to the year 2025. Dashed lines correspond to analysis using 15 years of data (1994–2008), dotted lines correspond to using 25 years of data (1984–2008), and solid lines correspond to using 20 years of data (1989–2008). Due to the similarity in results in some cases all lines are not clearly distinguishable from each other always.

(TIFF) Click here for additional data file.

Rates of yield change in kg/ha/year/year.

(TIFF) Click here for additional data file.

Year 2008 yields.

(TIFF) Click here for additional data file.

Similar to Figure 1 in the main text but only for the Brazilian Legal Amazon.

(TIFF) Click here for additional data file.

Country data source, number of political unitsterm·mth analyzed per country, time frame and number of official statistics collected per crop for the period 1989 to 2008.

(DOCX) Click here for additional data file.

Current yields, projections and production for the Brazilian Legal Amazon.

(DOCX) Click here for additional data file.

Comparison with the future U. S. crop yieldsterm·dom reported by the USDA-ERS [91] (maizeterm·dom, wheatterm·dom, and soybeanterm·dom were reported in bushels per acre, riceterm·dom in pounds per acres from USDA-ERS and converted to ton/ha), and global wheatterm·dom yields reported by the FAO-OECD [90].

(DOCX) Click here for additional data file.

Additional Data, Model Fitting, Rates, and Inter-comparison information.

(DOC) Click here for additional data file.

Rates observed for each crop and country.

(XLSX) Click here for additional data file.