1 Introduction

1.1 Crop yields in the 21st century

Plant productivity and crop yields are a function of environmental conditions (atmospheric CO2 concentration, soil properties, weather, pests) and management choices (e.g., planting date, irrigation, fertilizer application). Climate and environmental changes in the last few decades have reduced the yields of certain crops, including maize, wheat, and barley (Lobell et al. 2011; Lobell and Field 2007). Concurrent technological advances have produced new hybrids and cultivars, such as cold and drought-tolerant crops, allowing for earlier planting (Kucharik 2006), for example. Such advances have offset some climate-related losses (Kucharik 2008). Nevertheless, yield growth in general has slowed while yield variability has increased (Kucharik and Ramankutty 2005).

Crop yields are expected to continue changing in the 21st century in response to increasing atmospheric CO2 concentration, as well as to changes in soil water availability, growing season length, weather extremes, and management choices (Porter et al. 2014). Numerical models allow us to make crop yield projections into the future using statistical or mechanistic crop models. Leemans and Solomon (1993) made an early attempt at this and showed increased yields in the high latitudes due to warming and mixed results elsewhere mainly due to changes in water availability; they did not address the role of CO2 fertilization in their study.

McGrath and Lobell (2013) use a statistical model to estimate the effects of CO2 fertilization in the absence of climate change and find country-level yield increases up to 21 % for CO2 at 100 ppm over current levels. Tuberous crops are estimated to see the strongest yield increases, followed by C3 crops (e.g. soybean, wheat, and rice), and C4 crops (e.g. corn and sugarcane). Relative to C4 species, crops that use the C3 photosynthetic pathway experience enhanced photosynthesis under increased CO2 and see greater improvements in water use efficiency (WUE), leading to stronger improvements in overall yield. McGrath and Lobell also find that crop yields increase less for irrigated and more for arid rainfed crops because water-limited plants benefit more from enhanced WUE.

A majority of contemporary modeling studies show declining yields during the 21st century (Challinor et al. 2014). Rosenzweig et al. (2014) summarize 21st century yield projections from the Agricultural Model Intercomparison and Improvement Project (AgMIP) using mechanistic models of varying levels of complexity. Models simulating interactive soil nitrogen tend to simulate declining yields for RCP 8.5 (Representative Concentration Pathway in which anthropogenic emissions increase the climate forcing by 8.5 W m−2 by the year 2100 relative to 1850) at middle to low latitudes, especially in the absence of CO2 fertilization. They simulate increasing yields at high and some middle latitudes, mainly in areas with climates historically too cold to grow crops. On the other hand, models without interactive soil nitrogen tend to simulate more widespread yield increases because such models do not account for nitrogen as a dynamic factor that limits plant growth. Rosenzweig et al. refrain from quantifying their results in summary form due to the models’ wide range of outcomes; rather they emphasize the urgent need for additional crop modeling research.

PEGASUS, one of the models that participated in Rosenzweig et al. that includes interactive soil nitrogen and accounts for the application of nitrogen fertilizer, simulates similar trends as Rosenzweig et al. (2014) for a range of RCPs: yield increases in the middle and high latitudes with CO2 fertilization, mostly for soybean and spring wheat, sometimes even in areas where these crops already grow today; otherwise yield decreases that become more widespread without CO2 fertilization (Deryng et al. 2014). Deryng et al. experiment with a direct heat stress effect on simulated crop yield for when heat extremes co-occur with crop flowering and show further declines in crop yields. They suggest that, without CO2 fertilization, more than 80 % of the global average yield losses under RCP 8.5 may be avoided with RCP 2.6.

Using statistical relationships (generally these do not account for CO2 fertilization), Tai et al. (2014) calculate global crop yield losses of more than 10 % by 2050 from climate warming alone for RCP 4.5 and RCP 8.5. Tai et al. find that declining ozone pollution reduces yield losses in RCP 4.5, while exacerbated ozone pollution increases yield losses in RCP 8.5. Also using statistical relationships, Lobell and Tebaldi (2014) find the probability of 10 % yield-losses over the next 20 years (when food demand may peak under projected population trends) to have increased from 1-in-200 from internal variability alone to 1-in-10 for corn and 1-in-20 for wheat under RCP 8.5 climate trends. Urban et al. (2012) predict declining yields and increasing yield variability under the climate change projected by several GCMs (Global Climate Models). Schlenker and Roberts (2009) add that crops’ limited ability to adapt to extreme temperatures may contribute to yield declines.

1.2 Irrigation trends

Döll and Siebert (2002) calculate a 1995 global irrigation withdrawal rate (i.e. gross irrigation) of ~962 mm yr.−1 applied on more than 2.5 x 1012 m2 of area equipped for irrigation; they calculate net irrigation at ~428 mm yr.−1, this defined as the additional plant transpiration of irrigated versus rainfed plants. Two climate models under a scenario similar to the RCP 8.5 simulate increasing irrigation requirements in two-thirds of the global area equipped for irrigation during the 21st century due to increasing evapotranspiration from rising temperatures (Döll 2002). Döll’s irrigation model uses a simplified crop representation and no CO2 fertilization effects.

Valipour (2014) finds increasing net irrigation requirements (from 345 to 390 mm yr.−1 globally) and increasing area of crops equipped for irrigation (globally from 12 to 16 % of the total crop area) on every continent from 1997 to 2011. Valipour (2014) extrapolates potential changes in the area equipped for irrigation by 2035 and 2060 from historical relationships between relevant variables and finds largest increases in Africa.

With 21st century climate from six GCMs, Arnell (2004) calculates declining water availability around the Mediterranean, as well as in parts of Europe, central and South America, and southern Africa. Population trends imply that increasing numbers of people will experience water stress in the 21st century, sometimes even in areas with increasing water availability.

1.3 Goals for this study

CLMcrop is a mechanistic model embedded in a complex, well tested, and well documented earth system model. CLMcrop work to date focuses on model development and on effects of crops on the atmosphere and carbon cycle (Levis et al. 2012, 2014b; Drewniak et al. 2013, 2014; Chen et al. 2015). Past CLMcrop studies have not considered climate change impacts on yield and irrigation.

This paper contributes to the Benefits of Reducing Anthropogenic Climate changE project (BRACE; O’Neill and Gettelman, this issue), which focuses on characterizing the difference in climate impacts between RCPs 8.5 and 4.5 (van Vuuren et al. 2011). Here we perform CLMcrop simulations to examine the relative magnitude of climate impacts on crop yields and irrigation that human societies may face under these RCPs by the end of the 21st century.

Ren et al. (Avoided economic impacts of climate change on agriculture: Integrating a land surface model (CLM) with a global economic model (iPETS), Climatic Change, submitted) use the same CLMcrop yield outcomes as input to a global economic model of energy, land use, and agriculture to estimate economic effects on the agricultural sector. And, as an alternative to CLMcrop, Tebaldi and Lobell (2015) employ an empirical statistical model to investigate corn and wheat yields under the same RCPs. Tebaldi and Lobell generate statistical relationships that separate the effects of climate and CO2 fertilization on C3 and C4 crops. They also quantify the changing exposure of corn and wheat to extreme temperatures without, however, accounting for corresponding yield responses.

2 Methods

2.1 Model version

We use a post-4.5 version of the Community Land Model (CLM), the land component of the Community Earth System Model (CESM) (Hurrell et al. 2013). Compared to CLM4.5 (Oleson et al. 2013), which represented only temperate corn, soybean, and spring wheat-rye-barley (rye and barley represented exactly as spring wheat), the post-4.5 version also represents sugarcane, rice, cotton, tropical corn, and tropical soybean using the CLM4.5 parameterizations with modified parameter values as in Badger and Dirmeyer (2015). Specifically for sugarcane and tropical corn we use the CLM4.5 temperate corn functional form because all three are C4 plants; for tropical soybean we use the temperate soybean functional form; for rice and cotton we use wheat-rye-barley. Badger and Dirmeyer have since assumed that cotton more resembles soybean; this update is not included here. The new crop plant functional type (pft) parameter values were developed for the Amazon Basin (see Badger and Dirmeyer (2015) Table 1); planting date windows in the Northern Hemisphere are shifted here by six months.

Table 1 1986–27,005 CLMcrop area-weighted global annual yields (tonnes ha−1) and irrigation amounts (mm yr.−1) and their RCP 4.5 changes (RCP 4.5 2081–2100 minus 20thC 1986–2005); additional RCP 8.5 changes (RCP 8.5 2081–2100 minus RCP 4.5 2081–2100); statistically significant changes appear in bold face (95 % confidence level using a Student’s T-test); the last four columns display ratios of the RCP 8.5 total changes over the RCP 4.5 total changes, showing that the former sometimes more than double the latter
Full size table

We define irrigation requirement in CLMcrop as any water applied to a crop by the model’s irrigation algorithm. The latter initiates irrigation to crop areas appropriately equipped according to Portmann et al. (2010), when the photosynthesis is water-limited at 6 am local time even if water in the simulated runoff is insufficient (Oleson et al. 2013). The model computes the deficit between current and target soil moisture in the root zone above any frozen soil and adds such water through irrigation over a four-hour period. The target soil moisture content per soil layer is a weighted average of (1) the minimum soil moisture content that results in no water stress in that layer (weighted by 0.3) and (2) soil moisture at saturation in that layer (weighted by 0.7). The relative weighting was determined empirically using CLM4.0.

CLMcrop applies nitrogen fertilizer over the first 20 days of crop development from leaf emergence, at prescribed rates that vary by crop type according to North American practices; rates do not respond to simulated nitrogen deficits (Drewniak et al. 2013; Oleson et al. 2013).

2.2 Simulations

Key environmental conditions affecting CLMcrop plant productivity and, therefore, crop yields are climate, atmospheric CO2 concentration, and soil nitrogen content (Oleson et al. 2013). Our simulations are intended to isolate the effects of changing CO2 and climate on crop yields and the corresponding water requirements. We perform global CLMcrop simulations at 1.9° by 2.5° in latitude and longitude, not coupled to active atmosphere, ocean, or other models.

CLMcrop’s simulated carbon and nitrogen pools require thorough spin-up to initialize subsequent simulations. We spin up CLMcrop for 1050 years with 299.7 ppm constant CO2 (CCO2) corresponding to the 1910 concentration, and constant climate (CC) cycling 1901–1920 meteorology (52.5 times to complete 1050 years) from a 20th century CESM simulation contributed to the CMIP5 effort (Meehl et al. 2012). For the first 900 years we use CLM’s accelerated decomposition method (Koven et al. 2013) to approach steady state; soil carbon and nitrogen pools are slowest to equilibrate.

Following the spin-up, we perform three simulation sets:

  1. 1)

    20 th C using transient atmospheric CO2, nitrogen deposition, and CESM-simulated meteorology for 1901–2005; we initialize 20thC from the end of the spin-up and use meteorology from the CESM simulation used in the spin-up.

  2. 2)

    RCP45, RCP85: as 20thC but with initial conditions from the end of 20thC and meteorology for 2006–2100 from single ensemble members of the CESM RCP 4.5 and 8.5 simulations from the set contributed to the CMIP5 effort (Meehl et al. 2012). The particular RCP and 20th century CESM simulations were chosen for their high frequency meteorological output suited for driving CLM land-only simulations. We compare the last 20 years from RCP45 and RCP85 (2081–2100) and 20thC (1986–2005) to assess avoided yield and irrigation changes by choosing RCP 4.5 over 8.5.

  3. 3)

    RCP45CCO2, RCP85CCO2: as RCP45 and RCP85 but with 359.8 ppm constant CO2 (the 1995 value as mid-point in the 1986–2005 base period) to expose crops to 21st century climate but not CO2 fertilization. This allows for the possibility of no productivity enhancement in response to elevated CO2, which may occur due to plant acclimation to elevated CO2 (Leakey et al. 2009), due to ozone damage (Tai et al. 2014; Lombardozzi et al. 2015), or even due to changing crop nutrition under elevated CO2 (Myers et al. 2014), all processes not represented in these simulations.

We bracket crop yield responses to nitrogen-fertilized versus unfertilized conditions and rainfed versus irrigated conditions by representing fertilized, unfertilized, irrigated, and rainfed crops. In all our simulations we prescribe idealized vegetation cover so that all pfts may grow anywhere given favorable environmental conditions. During the post processing we weight the model output by observed or hypothetical land cover maps. Most results here are weighted by observed present-day crop-specific (e.g., irrigated corn, rainfed soybean, and so on) grid cell weights (Portmann et al. 2010); we indicate otherwise when we use different scaling. After the weighting we group crop categories (C3, C4, rainfed, irrigated) to draw broad conclusions about the yield and irrigation responses to changes in CO2 and climate.

We define “present day” as the last 20 years of simulation 20thC (1986–2005) and “end of the 21st century” as the last 20 years of the RCP simulations (2081–2100). The Supplementary Information includes the definition of crop yield and additional information about the simulations and results.

3 Results and discussion

3.1 Present day

Grid-cell by grid-cell simulated yield in CLMcrop, converted from tonnes ha−1 to country-level crop production in millions of tonnes, compares well with Food and Agriculture Organization (FAO) reported data (Supporting Figure 1a). Including nitrogen fertilizer and irrigation, the model overestimates production mainly in parts of Africa, but also parts of Asia and South America (Supporting Figure 1b), where fertilizers and irrigation are not widely used, yet. Further discussion on Supporting Figure 1 appears in the Supporting Materials.

CLMcrop’s simulation of corn, wheat, soybean, and rice yields (Supporting Figure 2) compares favorably to observational estimates by Monfreda et al. (2008) (Supporting Figure 3). The simulation with fertilizer typically overestimates yields in developing countries where CLMcrop overestimates fertilizer application.

Nitrogen fertilizer application rates agree best with the observed (Potter et al. 2010) in North America because model application rates represent North American practices. CLMcrop application is overestimated in Africa and South America versus underestimated in China and Europe where farmers typically apply less versus more fertilizer than North American farmers, respectively. Typical rates rarely exceed 10 gN m−2 (Supporting Figure 4).

CLMcrop generally underestimates irrigation relative to actual water withdrawal estimates of Sacks et al. (2009) (Supporting Figure 4). CLMcrop irrigates when plants are water-limited, while farmers often over-irrigate, so withdrawals tend to exceed requirements. Underestimation of at least an order of magnitude occurs in east Asia and Indonesia partly because CLMcrop does not represent flood irrigation for rice. Maximum simulated irrigation rates of ~400 mm yr.−1 occur in northern India in agreement with Sacks et al. (2009). CLMcrop’s globally averaged irrigation rate of 1022 mm yr.−1 compares well with the Döll and Siebert (2002) calculation of 962 mm yr.−1 for gross irrigation and a similar value by Shiklomanov (2000) to which the model’s irrigation algorithm was calibrated.

3.2 End-of-21st century yields

CLMcrop projects declining 21st-century global yields for C3 crops without CO2 fertilization and for rainfed C4 crops. Global yields increase for C3 crops with CO2 fertilization and for irrigated C4 crops. RCP 8.5 changes in yield exceed, sometimes more than doubling, RCP 4.5 changes relative to the present day. The global scale changes are small due to variability at the regional-scale, though in most cases statistically significant (Table 1).

Changes do remain consistent across scales with a small set of processes that explain them. The C4 increases occur in middle-high latitudes due to more favorable climate. The C3 increases are more widespread due to CO2 fertilization. Increases reach 3 t ha−1 locally (Supporting Figure 5). On the other hand, yield declines exceeding 3 t ha−1 stand out in parts of tropical and subtropical Africa and South America due to increasing plant respiration and decreasing soil moisture, both from rising temperatures (Supporting Figure 6).

Irrigation limits C4 losses mostly to 1 t ha−1 and reverses the losses in a few middle-high latitude areas. CO2 fertilization limits C4 rainfed yield losses because it improves plant WUE. C4 irrigated crops do not benefit from improved WUE because they are not water-limited. The improvement by CO2 fertilization is less than by irrigation.

Irrigation does not benefit C3 crops under CO2 fertilization due to competing effects: CO2 fertilization enhances C3 irrigated more than C3 rainfed photosynthesis, helped by unlimited water; meanwhile, irrigated C3 crops do not benefit from improved WUE, while the rainfed do; one process offsets the other. C3 yields generally increase (mostly by less than 1 t ha−1) with CO2 fertilization, which in this case enhances both WUE and photosynthesis. CO2 fertilization reduces or reverses losses more for the C3 rainfed than for the C3 irrigated crops because the latter do not benefit from improved WUE.

Most yield changes exceeding 1 t ha−1 and a few smaller ones are statistically significant at the 95 % confidence level according to a Student’s T-test (Fig. 1). The magnitudes of greater yield losses in the subtropics and lesser losses or small gains in the middle and high latitudes as a function of CO2 fertilization and irrigation agree with past results (e.g., Rosenzweig et al. 2014; Deryng et al. 2014; Porter et al. 2014; Parry et al. 2004). Tebaldi and Lobell (2015) also show declining yields for corn and wheat with smaller declines for wheat under CO2 fertilization.

Fig. 1
figure 1

CLMcrop simulated yield changes (tonnes ha−1) weighted by the present-day coverage of crops (Portmann et al. 2010). Color indicates statistically significant changes at the 95 % confidence level using a Student’s T-test. (a,b) RCP45CCO2 2081–2100 minus 20thC 1986–2005 for C3 crops and (c,d) for C4 crops; (e,f) RCP45 2081–2100 minus 20thC 1986–2005 for C3 crops and (g,h) for C4 crops

Full size image

The patterns and magnitudes of incremental change from 20thC to RCP45/RCP45CCO2 (Fig. 1) are similar as from RCP45/RCP45CCO2 to RCP85/RCP85CCO2 (Supporting Figure 7 and Levis et al. 2014a) and consistent with the global-average changes (Table 1). This similarity in incremental changes suggests that choosing RCP 4.5 may limit the magnitude of yield changes that society will confront to about half relative to RCP 8.5.

The yields and net primary production (NPP) display consistent patterns of change (Supporting Figure 8). Lower NPP is partly due to less available water simulated by CLMcrop in response to higher evapotranspiration due to warming, in agreement with models participating in recent intercomparisons (Seneviratne et al. 2013). Maximum declines in soil water availability appear in central America, around the Mediterranean, and in parts of Asia (Supporting Figure 5). Warming also leads to higher autotrophic respiration, as well as quicker advancement through crop growth phases that are GDD-based (GDD = growing degree day); both contribute to lower NPP values. Increasing CO2 improves C3 and C4 WUE and enhances C3 photosynthesis, so C3 NPP benefits more than C4 NPP from elevated CO2.

CLMcrop simulates similar, though muted, changes for unfertilized crops relative to crops fertilized with nitrogen fertilizer (Table 1).

3.3 End-of-21st century irrigation requirements

Global irrigation requirements increase the most for C4 crops without CO2 fertilization under RCP 8.5 (23 %) and the least for C4 crops with CO2 fertilization under RCP4.5 (3 %). RCP 8.5 changes in water requirements exceed, sometimes almost doubling, RCP 4.5 changes relative to the present day (Table 1).

CLMcrop irrigation requirements increase because simulated soil water availability declines almost everywhere. Exceptions include areas like northwestern USA in RCP 4.5 and midwestern USA in RCP 8.5 (Supporting Figure 9 and Fig. 2; the latter shows only statistically significant changes) where soil water availability increases. In northwestern USA, C3 irrigation requirements do decline, but C4 irrigation requirements increase due to a local increase in C4 crop productivity (and hence transpiration) apparent from increased yields. The opposite effect, where a reduction in crop productivity and transpiration more than compensate for the decline in soil water availability and result in reduced irrigation requirements, occurs in limited parts of central Asia in RCP 4.5 and more extensive parts of Asia including India in RCP 8.5.

Fig. 2
figure 2

CLMcrop simulated changes in irrigation demand (mm yr.−1) weighted by the present-day coverage of crops (Portmann et al. 2010). Color indicates statistically significant changes at the 95 % confidence level using a Student’s T-test. (a,b) RCP45CCO2 2081–2100 minus 20thC 1986–2005 for C3 and C4 crops, (c,d) RCP85CCO2 2081–2100 minus RCP45CCO2 2081–2100 for C3 and C4 crops, (e,f) RCP45 2081–2100 minus 20thC 1986–2005 for C3 and C4 crops, and (g,h) RCP85 2081–2100 minus RCP45 2081–2100 for C3 and C4 crops

Full size image

CO2 fertilization reduces irrigation requirements (Supporting Figure 9 and Fig. 2) as a result of improved WUE. C4 crops benefit from declining requirements more than C3 crops because the latter also experience enhanced photosynthesis under elevated CO2, which results in greater productivity and transpiration. This causes C3 crops to require more irrigation with CO2 fertilization on the global scale (Table 1).

3.4 End of the 21st century where crops do not exist currently

Bassu et al. (2014) hypothesize that high-latitude regions may become favorable to corn production with global warming. In areas where crops are not yet grown, we weight all CLM crop pfts equally in every grid cell and find that C4 crops respond to RCP 8.5 warming by growing in widespread boreal areas of the Northern Hemisphere. Yields increase the most (by more than 2 t ha−1) for the irrigated C4 crops in RCP85 relative to 20thC (Supporting Figure 10) and the least for the rainfed C4 crops in RCP85CCO2 relative to 20thC (not shown).

C3 yields increase by less than 2 t ha−1 in boreal regions because C3 crops are more cold tolerant and do not benefit from warmer and longer growing seasons as much as the C4. However, with CO2 fertilization C3 yields increase by 1–2 t ha−1 in Australia and the Sahara due to enhanced photosynthesis and improved WUE. In the Sahara the benefit is limited to the irrigated C3 crops because water is almost entirely unavailable to the rainfed crops.

All yield increases projected by the CLM may be overestimated due to lack of a direct heat damage parameterization.

4 Summary and conclusions

CLMcrop simulations project declining C3 yields without CO2 fertilization and declining C4 yields without irrigation for both RCP 4.5 and RCP 8.5 globally. Largest declines occur in tropical and subtropical Africa and South America. CLMcrop simulates increasing C3 yields with CO2 fertilization and C4 yields with irrigation globally. Increases tend to be smaller than the decreases. The C3 increases occur worldwide due to CO2 fertilization; the C4 increases are limited to the middle and high latitudes due to warming. These results are in line with the conclusions of the Intergovernmental Panel on Climate Change’s Fifth Assessment Report (Porter et al. 2014).

The RCP 8.5 changes in yield exceed, sometimes more than doubling, RCP 4.5 changes relative to the present day. Similarity in the patterns and magnitudes of incremental yield changes from the 20thC to the RCP 4.5 and the RCP 8.5 simulations suggests that choosing RCP 4.5 over RCP 8.5 may limit the magnitude of yield changes that society will confront to about half.

Most agricultural conversion currently occurs in the tropics and subtropics (Ciais et al. 2014) and CLMcrop simulates yield losses most prominently in such regions of Africa and South America. Irrigation generally mitigates the losses, so irrigation may play a key role in maintaining current yields in the future. The cooling and moistening effects of irrigation (Sacks et al. 2009) could limit losses further in coupled simulations that allow for land-atmosphere interactions and feedbacks, keeping in mind that irrigation may simultaneously weaken the land-atmosphere coupling (Badger and Dirmeyer 2015). In this work we do not consider potential challenges of expanding irrigation infrastructure to rainfed areas.

Similar to irrigation, CO2 fertilization also mitigates yield losses, though irrigated C4 crops benefit less because they lack the photosynthesis enhancement that C3 crops get and the WUE improvement because they have unlimited water. Such differences in C3 and C4 responses to irrigation and CO2 fertilization have been documented by Leakey (2009) and others.

In our simulations, crops equipped for irrigation require more water by the end of the 21st century, except in certain areas under CO2 fertilization especially in simulation RCP85. Such localized decreases are due to improved WUE with CO2 fertilization or to declining crop productivity with rising temperatures and, therefore, reduced transpiration. While CO2 fertilization may reduce water requirements, it may increase rainfed crop nitrogen requirements by locally enhancing runoff and leaching.

Global irrigation requirements increase most for C4 crops without CO2 fertilization under RCP 8.5. These simulations caution that society may face increasing irrigation requirements and declining yields under less than optimum CO2 fertilization (e.g., due to photosynthetic acclimation to elevated CO2) or under irrigation shortages. And as with the yield results, the changes in water requirements for RCP 8.5 exceed, sometimes almost doubling, RCP 4.5 changes relative to the present day.

Our simulations indicate that crops fertilized with nitrogen gain bigger gains and suffer greater losses than unfertilized crops. Despite the greater vulnerability of fertilized crops to climate and environmental change, fertilized crops deliver sufficiently higher yields than unfertilized crops making fertilizers (like irrigation) a potential element in supporting the world’s growing food demand.

Some scenarios of future agricultural production anticipate planting crops further north than now, as well as in some arid and semi-arid regions. Our results indicate that soil amendment, irrigation, and fertilizers would likely be necessary to make the effort worthwhile, thereby expanding the environmental footprint of agriculture through land degradation, increased water and fertilizer use, and the loss of wilderness. Not accounting for changes in irrigation infrastructure, Ramankutty et al. (2002) and Zabel et al. (2014) show maps of changing suitability of the land for cultivation; they agree with our result that boreal regions may become more favorable to agriculture under RCP 8.5 conditions. Still, agricultural intensification in areas of existing cultivation may help curb our need for agricultural expansion (Smith 2013).

Rosenzweig et al. (2014) found a wide range of yield outcomes across crop models (like Olesen et al. (2007)) that were not attributable to any specific methodology. Models of common lineage sometimes generated much different results, pointing to the urgent need for crop modeling research. Evans (2013) maintains that reliable crop models have yet to be developed. Areas of CLMcrop uncertainty and, therefore, potential development include:

  • The lack of direct heat damage on crop yield, such as the damage that occurs when heat waves co-occur with the flowering or grain-fill phase of certain crops (Deryng et al. 2014). Representation of heat damage is expected to reduce CLM yield projections (Teixeira et al. 2013), while irrigated crops are expected to emerge as less susceptible (Butler and Huybers 2015).

  • The extrapolation of Amazon Basin crops to pan-tropical crops and the potential misalignment between planting and local rainy seasons. CLMcrop buffers for this by selecting planting dates from planting windows. We acknowledge that sub-optimal sowing date selection relative to available soil moisture is a real (not just simulation) problem in the tropics (e.g., Paixão et al. 2014; Dzotsi et al. 2003).

  • Missing common practices such as double cropping, crop rotations, and tillage in the model. The lack of double cropping and the potential misalignment between planting and the rainy season may lead to underestimated tropical crop yields that are partly compensated by high fertilizer application rates.

  • The lack of spatially and temporally realistic fertilizer application to account for evolving practices as societies advance technologically.

  • The need to simulate winter crops (e.g., winter wheat).

  • The need for additional CLM simulations using meteorology from multiple CESM ensemble realizations per scenario, to quantify the uncertainty due to inter-ensemble variability. Despite such need, the results of the current study remain valid until resolution of higher order uncertainties in this list, as well as the uncertainty due to model spin-up.

Spin-up emerged from this study as a research topic. Here we choose to spin up CLMcrop’s carbon and nitrogen pools to equilibrium before initializing simulation 20thC (see Methods). We do not use realistic historical transient agricultural expansion, irrigation capacity, and fertilizer application because we do not have the capability, yet, with CLMcrop. Follow-up work may include such detail so as to evaluate the sensitivity of CLMcrop’s results to realistic introduction of agriculture, irrigation, and fertilizer application.