Predicting the impact of changing CO₂ on crop yields: some thoughts on food

I. Rising CO₂ and agricultural crop yields

Although terrestrial plants evolved at a time of high atmospheric CO₂ (4–5 times present values), concentrations appear to have declined to relatively low values during the last 25–30 million yr (Amthor, 1995; Bowes, 1996). However, records of atmospheric CO₂ concentration beginning in the late 1950s on Mauna Loa provided proof that the global atmospheric concentration of CO₂ was increasing (Keeling et al., 1976). These recent increases, and the projected concentrations of atmospheric CO₂ (i.e. 500–1000 µmol mol⁻¹ by the end of the 21st century; see Houghton et al., 2001), therefore represent an upsurge of an essential resource, exceeding anything plants have experienced since the late Tertiary (Pearson & Palmer, 2000; Crowley & Berner, 2001).

Although scientists have known for 200 yr that enhanced concentrations of CO₂ stimulate the growth of plants (T. De Sassure, 1804, as cited in Kimball, 1983), the number of studies involving CO₂ fumigation of agricultural and horticultural plants increased markedly in the 1960s and 1970s (Allen, 1979). By the early 1980s, Kimball (1983) had documented over 400 observations with respect to agricultural yield in response to elevated CO₂ conditions, data that could potentially be used to forecast the impact of rising CO₂ on crop productivity.

However, many of these observations involved the response of individual agronomic plants grown in pots in glasshouse or growth cabinet conditions. It seems unlikely that the response of single plants to CO₂ could act as a reasonable surrogate for prediction of the future of global agriculture given the spatial and temporal aspects of plant function, from the molecular to the ecosystem level. In addition, CO₂ is not just the source of carbon in the biosphere, but a longwave-radiation trapping gas, with consequences for surface temperature and precipitation, climatic variables that affect crop productivity.

II. Expanding methodologies

Clearly, a better and more thorough analysis of how plant growth and yield would respond to CO₂/climate was needed. To that end, a number of methodological papers throughout the 1980s and early 1990s described new, innovative means to simulate future CO₂/temperature/climate such as soil–plant–atmosphere research (SPAR) units (e.g. Mortensen, 1982), temperature gradient tunnels (TGT; e.g. Horie et al., 1991), open-top chambers (OTC; e.g. Rogers et al., 1983; Drake et al., 1989) and free-air CO₂ enrichment (FACE) systems (e.g. Allen et al., 1985; Hileman et al., 1992a). The FACE technology was developed, in large part, because earlier results using enclosures (laboratory and field) indicated that small plot size and artificial enclosure could produce microclimatic and edge effects that would influence crop responses (Hendrey, 1994). However, as with all simulations, each methodology is able to add (or subtract) various abiotic parameters from the system (Table 1). In general, as the spatial/temporal scale increases, microclimatic effects do diminish, but short-term control of variables becomes more difficult. For example, a 5-s sample of air from an elevated CO₂ OTC was within ± 10% of the set point 80% of the time (L. H. Ziska, unpublished data; Hileman et al., 1992b), whereas a 5-s air sample from an elevated rice (Oryza sativa) FACE system in Japan was within ±10% of the set point only 50% of the time (Table 2; Okada et al., 2001).

III. Which methodology gives the ‘truest’ prediction of future yields?

As systems for simulating future atmospheric composition and/or climates have become more sophisticated over greater spatial scales, is it fair to state that the results from such newer systems represent (or are closer to representing) the ‘true’ response of crop yields to projected changes in CO₂/climate? In a recent hypothesis first published in the Philosophical Transactions of the Royal Society (Long et al., 2005), and later reiterated in Science (Long et al., 2006), Long and colleagues argued that current modeling efforts overestimate the impact of increasing CO₂ on future crop yields, because the models used are parametrized with data obtained from earlier ‘enclosure studies’ and not from the more sophisticated FACE systems.

However, Long et al.'s (2005, 2006) hypothesis made no distinction among ‘enclosure studies’ which would include crop yield data obtained from a variety of sources, including single plant experiments in growth chambers or glasshouses, as well as assemblages of plants grown in field-based open-top chambers. As we have already pointed out, single plant data are unlikely to reflect global crop responses to CO₂ and/or climatic change; hence, the comparison of ‘enclosure studies’in toto with FACE systems requires further elucidation as to what exactly is meant by the term ‘enclosure studies’ and whether, in fact, all enclosure studies overestimate the impact of rising atmospheric CO₂ on crop yields.

For a more precise comparison, we examined relative yield response to elevated CO₂ for a number of enclosure methodologies for three crops of global significance: rice, soybean (Glycine max) and wheat (Triticum aestivum). In examining the literature we have defined ‘experiment’ as a single replicated study of one cultivar, i.e. multiple years or cultivars are multiple ‘experiments’ (cf. Ainsworth et al., 2002). In addition, as most comparisons of yield enhancement (relative to ambient, background concentration) have been made at ∼700, rather than 550 µmol mol⁻¹ CO₂ (i.e. the concentration used in the FACE experiments), it is easier to use 700 µmol mol⁻¹ as a common point of comparison. It should also be stressed that ambient CO₂ concentration itself has changed appreciably during the experimental era (e.g. from ∼320 in early enhancement studies to a current concentration of ∼380 µmol mol⁻¹), such that we have scaled the relative yield to a ratio of 700 to 370 µmol mol⁻¹ CO₂ (Tables 2–4). (The use of 370 rather than 380 µmol mol⁻¹ is a consequence, in part, of the fact that many recent papers still list ambient CO₂ concentration as 370 µmol mol⁻¹; e.g. Morgan et al., 2005.) Scaling the observed (published) response to these concentrations was accomplished using a beta (β) factor (β = [(Y_H – Y_L)/Y_L]/ln(C_H/C_L), where C_L is the ‘low’ (always approximately ambient; 315–408 ppm) CO₂ concentration used, C_H is an elevated CO₂ concentration used, Y_L is yield in the C_L treatment, and Y_H is yield in the C_H treatment for any published experiment). Following calculation of the beta factor for a given experiment, the relative stimulation of yield (1 + [(Y_H – Y_L)/Y_L]) was then solved for 700 and 370 µmol mol⁻¹ as C_H and C_L, respectively (Tables 2–4). The appropriateness of using a beta response function for these species is consistent with published yield response curves over a range of CO₂ values (e.g. fig. 1 in Baker et al., 1996 for rice; fig. 14 in Baker & Allen, 1993 for soybean). In addition, we have only included data for plants under favorable conditions of temperature, water, light, nutrients, and unmanipulated ozone concentrations. There are obvious complications imposed by both abiotic and biotic interactions with regard to stimulation of crop yields by elevated CO₂, but these will be discussed in greater detail in the latter part of this review.

IV. Comparing responses among methodologies

V. Methodology vs future uncertainty

In the methodology debate, there is another, larger, consideration. The goal of any of the methodologies discussed so far is to quantify how crop yields will respond to projected changes in CO₂ and associated changes in climate in order to assess the potential threat to future food security. Suppose that, in a side-by-side comparison between methods, that one of the methods (call it ‘A’) could be shown to be the superior technology for simulating future CO₂ concentrations. Would the response of a given crop cultivar to a single elevated CO₂ concentration provide an accurate picture of future yield? Probably not. It is still unclear if smaller increases in CO₂ concentration provide a similar response to that found in larger ‘doubling’ experiments, and it is likely that different varieties, probably with different responses to a change in CO₂ concentration, will be grown in the future.

But is any quantification of yield in response to increasing CO₂ by itself sufficient to provide us with an accurate estimate of future yields? It is becoming increasingly clear that any effect of CO₂ concentration on yield will be modulated by other abiotic variables such as temperature, moisture, and nutrient availability. In addition, biotic factors such as weeds, insects, and diseases might modify the effect of increasing CO₂ concentration on yield. Indeed, the large range of responses obtained with all methodologies, as documented in Tables 2–4, for experiments under nominally optimal conditions suggests that large abiotic and/or biotic interactions exist. Both abiotic and biotic factors, in turn, are also likely to be directly or indirectly altered by changes in CO₂ and/or climate.

Therefore, if our goal is to achieve greater consistency between experimental and model results for crop yield in a future climate, then an improved understanding of how abiotic and biotic parameters alter the response of crop yield to CO₂ concentrations may also be crucial in providing a more accurate estimate of agricultural productivity and potential threats to food security. For the remainder of this review, we would like to focus on what we perceive as key areas of biotic and abiotic uncertainty; and to address, in a preliminary fashion, how questions associated with such uncertainties could be used to improve yield forecasts of rice, soybean, and wheat with increasing CO₂ concentration.

VI. Abiotic uncertainties

Two of the most important edaphic factors related to crop yields are soil moisture and nutrient availability. For rice and wheat, there are now sufficient data to indicate that any stimulation of crop yield by increasing CO₂ will be dependent on nitrogen (N) availability (e.g. see Kim et al., 2003 for rice; Wolf, 1996 for wheat) whereas stimulation of yield by CO₂ in soybean appears to be independent of supplemental N (Cure et al., 1988a). However, a number of questions remain unaddressed. What is the optimal application of N needed with increasing CO₂ to maximize rice and wheat yields? If root growth is stimulated by CO₂, how will this affect the timing of N application and temporal exploitation of soil resources? If, for economic reasons, N is limiting for a given region, what will be the expected impact on crop yield as CO₂ rises? What about the supply of other nutrients, particularly phosphorous?

In contrast to nutrients, there is a surfeit of data indicating that, under water-limiting conditions, the indirect effect of CO₂ on stomatal aperture (and potential reductions in transpirational water use) may enhance the relative effect of elevated CO₂ on crops (see Polley, 2002 for a review). However, here too there are a number of unresolved questions. How will elevated CO₂ affect crop yields under flooded conditions? Will yield sensitivity to water quality (e.g. salinity) be similar at higher CO₂ concentrations to the yield sensitivity today?

One of the largest uncertainties remains air temperature. Initial assessments based on photosynthetic biochemistry suggested a positive interaction between projected increases in temperature and CO₂ (e.g. Long, 1991). However, such positive interactions do not necessarily translate into additional seed yield with simultaneous increases in CO₂ and temperature as opposed to elevated CO₂ alone (e.g. Matsui et al., 1997; Batts et al., 1998; Amthor, 2001). For rice, increasing CO₂ and temperature may, in fact, negate any yield enhancement related to increasing CO₂ (Moya et al., 1998; Ziska et al., 1996) as a result, in part, of reductions in transpirational cooling, higher canopy temperatures and increased pollen sterility (Fig. 1). The interaction of temperature and CO₂ seems more complex for wheat, with a majority of experiments indicating a reduction in yield with elevated CO₂ in combination with warming compared with elevated CO₂ alone (table 7 in Amthor, 2001). A few studies have compared responses of soybean crops grown for the full season under different temperature regimes (Sionit et al., 1987; Baker et al., 1989; Boote et al., 2005; Heinemann et al., 2006) but have found only minor effects of these treatments on the yield response to elevated CO₂.

Another atmospheric uncertainty is anthropogenic pollution. Ozone concentrations have certainly risen precipitously throughout the 20th century (Marenco et al., 1994), although some evaluations indicate that ozone may be declining, particularly in the eastern states (Fiore et al., 1998; EPA, 2004). Because ozone is phytotoxic, a number of studies have examined whether the increase in ozone will offset any positive effect of CO₂ on crop yield. In a review of studies of wheat, Amthor (2001) evaluated experimental protocols and concluded that four studies (all OTC experiments) were relevant with respect to a combination of possible future ozone and CO₂ concentrations. Recent work with soybean has indicated that ozone fluxes that suppressed net photosynthesis, growth and yield of soybean were generally much less harmful if plants were grown simultaneously with elevated CO₂ (Booker & Fiscus, 2005). Heagle et al. (1998) found that increasing ozone concentration above the ambient outdoor concentration increased the seed yield response to elevated CO₂. Although Heagle et al. (1999) also reported a significantly smaller yield response to enhanced CO₂ at low ozone, it is unlikely that high ozone is necessary for CO₂ stimulation of yield; otherwise, there would be little CO₂ response of soybean in SPAR, glasshouse or growth chamber experiments. The fact that consistent responses to CO₂ were observed for a wide range of methodologies, some of which filter ozone, indicates that ambient ozone concentrations per se may not impose a significant limitation to CO₂ enhancement of yield in soybean, contrary to some reports (e.g. Long et al., 2005). In contrast to soybean and wheat, almost no data have been published for ozone, CO₂ and rice yields. Overall, given the spatial and temporal distribution of ozone, and the fact that different stages of plant growth will exhibit different sensitivities, a satisfactory understanding of how ozone could modify crop yield response to rising CO₂ has not been forthcoming, and should be a priority for further research.

VII. Biotic uncertainties

Since the inception of agriculture, insects and diseases have limited crop yield potential. Will such limitations increase or decrease in response to future changes in CO₂/climate? Certainly it is reasonable to expect that climate stability with respect to temperature and precipitation is likely to affect the range of specific species of insects and diseases for a given crop growing region. For example, Cannon (1998) has suggested that migratory insects could colonize crops over a larger range in response to temperature increases, with subsequent reductions in yield. Gutierrez (2000) has suggested that predator and insect herbivores are likely to respond differently to increasing temperature, with possible reductions in insect predation (i.e. greater insect numbers). Unfortunately, while there is evidence suggesting that insect damage could increase as a function of climate, specific experimental results related to rice, soybean and wheat remain scarce. Similarly, while we recognize plant–pathogen interactions as a factor affecting crop yields, our ability to predict CO₂/climate change impacts on pathogen biology and the impact of subsequent changes on the yield of rice, soybean or wheat is tenuous at best as specific experimental data are not available.

Given the importance of weeds to crop production, it is surprising to find so few assessments of how changes in CO₂/climate will alter their impact on agriculture (a recent review of all crop/weed competition studies is given in table 2.2 in Ziska & Bunce, 2006). Yet, we are aware of only a handful of weed/crop competition studies with respect to soybean (Ziska, 2000; Ziska & Goins, 2006), one study with respect to rice (Alberto et al., 1996) and no studies with respect to wheat, where the effects of projected changes in CO₂/climate on seed yield have been quantified (Fig. 2). As with pests and diseases, CO₂/climate effects on weed biology and crop/weed competition represent a significant biological uncertainty with respect to predicting future yields of rice, soybean and wheat.

Another biological uncertainty is the range of intraspecific variability within a given crop to changes in CO₂ and/or climate. Morgan et al. (2005) suggested that variation among soybean cultivars in yield response to enhanced CO₂ was small and unlikely to be a factor in soybean response in a FACE system. This is in contrast to cultivar comparisons that have shown large differences in response ratios in glasshouse and OTC studies for soybean (Heagle et al., 1998; Ziska & Bunce, 2000; Ziska et al., 2001). Overall, in fact, there are at least enough preliminary data to suggest that genotypic variation in the response of rice, soybean and wheat yields to CO₂ could be substantial (e.g. Manderscheid & Weigel, 1997; Moya et al., 1998; Ziska et al., 2001, 2004; for a recent review, see Newton & Edwards, 2006). Analysis of cultivar differences in CO₂ responsiveness and exploitation of such differences could significantly improve global food security as CO₂ increases. Much of the US Department of Agriculture, Agricultural Research Service (USDA-ARS) effort for soybean in this regard was switched to the SoyFACE system at its inception, because that system is better suited to field comparisons among multiple cultivars, but no reports of comparative yield responses from FACE have been published to date.

However, to properly assess intraspecific variation, other variables in addition to CO₂ need to be examined. These data are noticeably lacking, particularly for soybean and wheat. In a glasshouse study involving 17 different rice varieties, genotypic variation in response to CO₂ was negated as temperature increased, because, for all cultivars, high temperature resulted in pollen sterility no matter the CO₂ concentration (Ziska et al., 1996). Similarly, for a field OTC study, the combination of increasing CO₂ and air temperature resulted in reduced grain yield and declining harvest index compared with increased CO₂ alone for three rice cultivars (Moya et al., 1998). Still, given that there are over 100 000 rice cultivars (and thousands of soybean and wheat cultivars), it seems reasonable to suggest that a more thorough evaluation of the response of crop germplasm to CO₂/climate is warranted. Unfortunately, at present, we are unaware of any such systematic evaluation in regard to CO₂/climate at the government, university or corporate level. Yet such an assessment is imperative for identifying those cultivars that could maintain, or improve, yields in response to changes in CO₂/climate and improve food security.

VIII. Uncertainty vs methodology

The uncertainties listed here are not meant to be inclusive, but illustrative. If our goal is to improve our predictive capacity regarding crop yields with respect to CO₂/climate, then a better understanding/quantification of these (and other) uncertainties would seem to be a high priority. Which methodology, then, is best suited to studying these uncertainties? At present, no one methodology is capable of providing a perfect evaluation of future CO₂/climate with precise control (i.e. all current methodologies have limitations). However, given that CO₂/climate is likely to affect a range of biological properties over spatial and temporal scales, a number of methodologies could, and should, be used to improve our knowledge of these uncertainties. For example, superambient changes in CO₂ for a range of temperatures could be assessed for crop yields using SPAR units; greenhouses or growth chambers could be used to assess genetic and proteomic differences and responses among crop lines to CO₂/climate; FACE could be used to assess ecosystem interactions among weeds, pests and diseases, etc. Ultimately, appropriate technologies should be determined by the specific level(s) of organization the researcher wishes to investigate and/or the relative ability to adequately control the environmental variable of interest.

This does not mean that we should ignore methodological deficiencies. New and innovative strategies to simulate and control a range of projected environments using methodologies that include control of abiotic/biotic uncertainties over large spatial scales are crucial in bettering our understanding of the underlying plant processes likely to be affected by projected changes in CO₂/climate. Understanding of such processes, in turn, will allow us to extrapolate experimental data to improve model scenarios for a wide range of crop yields with respect to global climate change. To that end, improvement of current systems, such as inclusion of heaters to allow warming to be included as an experimental variable in large FACE rings, and improved technology to minimize microclimate effects in SPAR units and OTCs, as well as new approaches (e.g. urbanization as a surrogate for climate change; see Ziska et al., 2003), need to be explored and characterized.

IX. Modelers and experimentalists

Given the unprecedented scale and scope of anticipated changes in CO₂/climate, it is tempting to ignore the large number of uncertainties as being unquantifiable, and to focus on one or two parameters (e.g. CO₂ in isolation from other changes) that can be well defined. In general, such an approach is consistent with that of many experimentalists who work in large part on quantifying one or two variables, usually in the short term (1–2 yr). Such an approach is also certainly valid as it provides key information, but only for a specific set of circumstances.

However, given the large adjustment in scale, uncertainties in crop yields may be magnified in global change models. This is not an expression of the modeler's aptitude; rather, it is a reflection of experimental unknowns. For example, early projections by Kropff et al. (1993) of the effect of climate on rice yields assumed that rising CO₂ and temperature acted independently (i.e. the response to CO₂ did not change with temperature). Yet, since these early assessments, it has become apparent that elevated CO₂ may exacerbate the negative effects of temperature, particularly on pollen sterility (Kim et al., 1996; Matsui et al., 1997), and that higher night-time temperature may limit rice yields (Peng et al., 2004). Incorporation of temperature/CO₂ interactions might, of course, improve the efficacy of model projections regarding the consequences of CO₂/climate changes on rice yield (as might incorporation of weeds, pests and disease impacts, or cultivar variability, or nitrogen deposition, etc.).

X. Final thoughts

Given a current population of more than six billion with a projected increase of an additional billion every 12 yr, being able to reliably predict the impact of changing CO₂/climate on global crop productivity certainly should be ‘food for thought’ for scientists and policy makers (Long et al., 2006). How, then, do we improve our predictive capacity? Long et al. (2005, 2006) contended that re-evaluation of crop yields in response to elevated CO₂ concentration using FACE technology is a key means to improve model reliability. While FACE technology can offer some advantages to the experimentalist, we have also shown that the relative effect of CO₂ on yield stimulation of rice, soybean and wheat is, with the exception of single plant evaluations, consistent across a range of CO₂ concentration control methodologies.

Given the importance of models in predicting the impact of CO₂ and/or climate change on food security, it is of obvious interest to try, whenever possible, to improve the efficacy of existing models (Tubiello et al., 2007). However, in addition to the concerns related to methodology (Long et al., 2005, 2006), we would suggest a broader evaluation and inclusion of more of the biotic and abiotic factors mentioned here into any integrated assessment of rising CO₂ and crop yields. Incorporation of these factors will certainly require unprecedented cooperation and data sharing between modelers and experimentalists of all disciplines at the government, university and private sector levels – a daunting task in an age of specialization and increased competition for diminished resources. Nevertheless, by more precisely defining the impact of CO₂/climate on crop yield, we hope to be able to reduce uncertainty regarding global food supply in an uncertain and changing environment.