Will Climate Change Really Take Away Our Breakfast?

According to a recent high-profile Nature paper, climate change will make us go hungry. At least that’s what media reports on the study would have you believe. An article on the study in Vox asserts that “climate change will worsen hunger,” before intoning that “even America’s richest farmland can’t outrun climate collapse.” The same article was republished by Wired, and similar coverage appeared in CNN and The Hill.

The study’s authors have not done much to dampen the hysteria around their research. Lead author Andrew Hultgren and co-author Solomon Hsiang have both, in various venues, warned that climate change is going to take away the most important meal of the day. They come to this conclusion on the basis of their finding that each degree Celsius of warming is associated with the loss of about 120 less calories per person per day. Extrapolate to a world that has warmed by 3C and it’s “like everyone giving up breakfast,” Hsiang tells The Hill.

Giving up breakfast (albeit a light one) sounds bad, but is that actually what the study is saying will happen? A closer inspection of the paper reveals it is not. The yield losses projected by the study are relative to a counterfactual world with no future climate change, not to the yield levels of today. The casual reader would be forgiven for missing this important point: it is not mentioned in the abstract, and only alluded to in the caption to a figure. Comparing yields (or deaths, or dollars) to a counterfactual is very common in climate impacts literature—perhaps the authors considered this aspect of their study design so unremarkable it didn’t merit mentioning. It is harder to understand, however, why they portrayed the losses as absolute rather than relative in their interviews with the media (and by the same token, why reporters who should be familiar with this kind of study accepted this interpretation).

As we have emphasized in earlier critiques of crop yield impacts literature, when we are discussing relative impacts it is crucial to contextualize them in terms of the baseline trend. For example, the paper’s central estimate for global average yield reductions under RCP 8.5 warming relative to the no-climate change counterfactual is 15.6%. This sounds not great until you remember that background yield trends are very great. Thanks to the Green Revolution, average global cereal yields—which include wheat, maize, rice and other staple crops—have increased over 215% since 1961, and this year’s output of key commodity crops is projected by the FAO to be the largest on record.

While Hultgren et al. do not project how much yields would grow in their counterfactual of a world where climate change stops in 2025—making their findings of declining yields in the face of climate change seem grander—a simple extrapolation of the last six decades of global yield growth provides a decent stand in. If yields did grow at the same rate until 2098, global agricultural yields would almost double from just over 4 tonnes/hectare in 2023, to just under 8 tonnes/hectare by the end of the century. A 15.6% decrease compared to this growth would still mean more than a 50% growth in global average yields. The authors’ projected decrease of 5.9% under the far more plausible RCP 4.5 warming scenario, meanwhile, would still mean a 72% increase in yields from today.

Contextualizing yield impacts can go a long way toward reducing hysteria around the estimates; but there’s also good reason to question the projections themselves. They are, by the authors’ own admission, extremely uncertain. According to the paper, the 5th percentile simulation for RCP 8.5 warming by end of century saw a 45.7% decrease relative to the no climate change counterfactual, while the 95th percentile simulation saw a 41 percent increase. In other words: according to these simulations, there is a moderate chance climate change could have a positive effect on yields. The authors put this chance at 47.7% for rice, 25.8% for maize, 17.9% for cassava, 32.9% for sorghum, and 18.5% for soybean under a high emissions scenario. The only crop for which a positive impact is very unlikely is wheat (3.8%). Generally, when the range of uncertainty for an estimate encompasses both positive and negative effects, it is not an estimate we should put much stock in.

The authors, to their credit, do account for the potential effects of farmer adaptation and CO2 fertilization in the estimates discussed above. But even with these important factors incorporated, their projections for yield change impacts are likely far too pessimistic. There are several reasons for this.

The first and most important is that their projections do not allow for further technological innovation. The authors’ model, which does mark a sophisticated upgrade on pre-existing methods to estimate agricultural producer adaptation response to climate impacts—in past models, farmers either responded perfectly or not at all—is nevertheless merely trained on historical data. These data include variables for country fixed effects and time trends, which are necessary to isolate adaptation responses to weather. In comparing relative yields to a counterfactual, however, these time and country terms get subtracted out—which is a problem because time trends and country fixed effects are exactly the terms that absorb the effects of technological innovation in a statistical model. Their projections, in other words, reflect the impact of climate change on crop yields in a world where farmers only have access to today’s technology. This doesn’t make the model completely useless: many farmers around the world use little modern agricultural technological technology, so this model can help us understand how variables like income may influence the diffusion of known technologies. In reality, though, it is very unlikely that farmers in 2098 will still be stuck with today’s agricultural technology, given ongoing research and development. Indeed, this exact point was made by one of the paper’s peer reviewers, who warned about people taking the authors’ estimates “at face value as a projection of the future.”

Some may assume there is little room to significantly increase crop yields beyond today’s levels—after all, the Green Revolution’s biggest gains are behind us, and farmers now harvest many times more than their grandparents did. But that view overlooks how future technological progress may differ substantially from the past. Advances in biology and genetics—including gene editing, genomic selection, and biological inputs like biopesticides—are enabling faster, more targeted improvements in seeds and crop management. Though they warrant more government support than they’re currently receiving, these innovations may allow producers to adapt to climate change in ways not captured by historical data, and not included in this study’s projections.

Beyond the problem of inability to account for future technological innovation, there is the issue of climate and socioeconomic scenarios. The author’s headline findings mostly come from their modelled projections under RCP 8.5, a scenario that is now thought to be very unlikely. Moreover, the authors pair RCP 8.5 with the pessimistic SSP-3 population and income scenario—a pairing that does not exist in the IPCC database, and is moreover physically impossible, as the incomes and population under SSP-3 do not produce enough energy demand to result in RCP 8.5 levels of warming. It also does not make much sense that the authors’ counterfactual assumes incomes grow at the same rate in a world where climate change suddenly stops as one in which it continues. It is hard to imagine the world—especially the developing world—seeing no economic penalty from suddenly ceasing to burn fossil fuels.

But where does breakfast come into this? Potential yields do not automatically translate into calories, because the amount of food that actually gets produced has to do with demand on the world market and producers’ ability to meet that demand. Markets may respond to climate change in numerous ways. If climate change is going to reduce expected yields, producers may be motivated by higher prices resulting from reduced supply to convert new land to agriculture or to plant more intensively on already cropped land. If land currently devoted to a given crop becomes unsuitable for that crop, producers may switch the crop they are planting or migrate the location of their crop.

But Hultgren et al. do not account for any of these market responses in their projection of the loss of breakfast. Their analysis, therefore, while maybe helpful for understanding the purely theoretical effects of climate change on production, does not do much to help us understand how much will get produced in the real world. As Professor Erin Coughlan de Perez points out in the Vox article, people switch crops or convert new land all the time—in the US, corn and soy production have already started migrating northward. The effect of crop switching and land conversion is not trivial. The authors themselves note most studies find these strategies and shifts in international trade reduce welfare losses by around 55%, and use this adjustment in their valuation of the social cost of carbon.

Looking beyond the headlines and hysteria, Hultgren et al. give us a very careful, data-driven endeavor that credibly quantifies how past producers have adapted to climate change with the technology that already exists. And it paints a plausible picture of the headwinds to maintaining historical yield growth under climate change. Indeed, the authors recognize exactly this in the abstract when they write that their results “indicate a scale of innovation, cropland expansion or further adaptation that might be necessary to ensure food security in a changing climate.” It’s just too bad that this is not the message they chose to communicate to the media.