FACE‐ing the facts: inconsistencies and interdependence among field, chamber and modeling studies of elevated [CO2] impacts on crop yield and food supply

Abstract

Rising atmospheric carbon dioxide concentration ([CO2]) is just one aspect of global climate change. However, it is important because it consistently stimulates the growth and harvestable grain production of C3 crops (Kimball et al., 2002; Long et al., 2004; Nowak et al., 2004; Ainsworth & Long, 2005), as well as benefiting C4 crops under drought stress (Ottman et al., 2001; Leakey et al., 2004, 2006). Meanwhile, high temperatures, drought stress and rising ozone concentrations all have negative impacts on crop production (Gitay et al., 2001; Parry et al., 2004). Furthermore, rising [CO2] is unique in being globally almost uniform and so denying spatial proxies for temporal trends. As a result, Parry et al. (2004) singled out its effect as the largest uncertainty in projecting future global food supply. Free-air CO2 enrichment (FACE) experiments currently provide the most realistic measure of the future impact of elevated [CO2] on crop yields. Free-air CO2 enrichment experiments differ from enclosure studies in two salient respects: (1) they are conducted in the open air in farm fields without limiting growing space, or altering microclimate, precipitation or pest/pathogen access; and (2) the scale of the experiments is large enough to be comparable to agronomic trials (typically plots > 300 m2 compared with < 4 m2 in the case of enclosure studies). In Long et al. (2006) and Ainsworth (2008), we reported that stimulation of seed yield in response to elevated [CO2] is lower in FACE experiments than in enclosure studies of the world's four most important food crops. We suggested that the implications of this finding were as follows: modeling studies using CO2 fertilization factors derived from enclosure experiments may have overestimated future food supply; and additional field experiments are needed to understand in greater detail the mechanism of response and to drive research and development efforts to improve crop yields under future climatic conditions. This was argued with the proviso that FACE experiments have been limited in number and geographic range, as well as in the extent to which they currently investigate the interactive effects of the numerous elements of global change. These findings were vigorously challenged and then subsequently dismissed by Tubiello et al. (2007a,b) who argued that the findings of Long et al. (2006) were incorrect for three reasons: a statistically significant difference between FACE and nonFACE data was not adequately tested or proven; results from most crop model simulations are consistent with values from FACE experiments; and lower crop responses to elevated CO2 of the magnitudes in question would not significantly alter projections of world food supply (Tubiello et al., 2007a). This conclusion is in contrast to the findings of Parry et al. (2004), who with similar modeling approaches concluded that whether global food supply remained stable or declined with global change would depend critically on the response of the world's key grain crops to rising [CO2]. The consequences for human welfare will probably be severe if we underestimate the threat of global change to future food supply. This is all the more important given that long lead times will be necessary to produce germplasm better adapted to future growing conditions; typically it may take decades to identify adapted germplasm and then bring it to market in sufficient quantity for these major crops. Consequently, there is an immediate need to evaluate the currently available data correctly and use this to identify the best approaches to predict future food availability and to elucidate the mechanisms of crop responses to elevated [CO2] in order to generate improved germplasm. The statistical validity of comparing the mean response of C3 crops to elevated [CO2] from FACE experiments (Supplementary material Table S1) against the modeled best-fit response of chamber experiments (Long et al., 2006: fig. 2) was criticized because curve-fitting methods and data-pooling choices can bias fair comparisons (Tubiello et al., 2007a). In fact, Tubiello et al. (2007a) recommend that a better approach is to 'whenever possible, use the observed data – rather than "predicted" from curves lacking full biophysical explanatory power.' We agree, and given the number of C3 crop studies, it is possible to take a direct approach and limit the comparison of FACE experiments and chamber studies to those with similar ambient [CO2] and similar elevated [CO2] (Supplementary material Tables S1 and S2). The FACE data had an average ambient [CO2] of 367 ppm and an elevated [CO2] of 583 ppm, and were normally distributed with a mean yield response ratio of 1.14 (Fig. 1). The chamber data had an average ambient [CO2] of 373 ppm and an elevated [CO2] of 565 ppm, and a mean yield response ratio of 1.31 (Fig. 1). In contrast to the FACE data, the chamber data were not normally distributed (Shapiro–Wilks P < 0.001) and had a much broader range of