Complex Models Now Gauge the Impact of Climate Change on Global Food Production. The Results Are ‘Alarming’

Inside dozens of bankers boxes, stacked high in a storage locker in New York City, Cynthia Rosenzweig has stashed the work of decades: Legal pads covered in blue-inked cursive with doodles in the margins, file folders marked “potato,” graph paper with notations of rainfall in Nebraska and Kansas.

Rosenzweig has worked at NASA’s Goddard Institute for Space Studies (GISS) at Columbia University since the 1980s, when researchers were delving deeper into the growing science demonstrating that human activity is warming the planet. But as her colleagues were focused on fossil fuel use or the impact of global warming on sea level rise, Rosenzweig, an agronomist by training, started to wonder what the changing climate would do to crops.

Among those thousands of stashed pages are copies of her first published work, a seemingly esoteric paper showing that higher levels of carbon dioxide would push wheat growing regions in North America farther north. The paper made Rosenzweig a pioneer, one of the first researchers to use simulation models to look at the specific impacts of climate change on agriculture.

At the time, the idea of climate change was just emerging in the public consciousness. Rosenzweig’s boss at NASA, James Hansen, was about to tell Congress, in seminal testimony, about the looming perils of the warming planet. With characteristic pluck, Rosenzweig sent Hansen a note before the hearing, telling him that scientists were overlooking a big potential threat. 

“We needed to model CO2 and precipitation,” she recalled in a recent interview. “We needed to understand the full impact of climate change on food.”

At the time, there was scant research on the intersection of climate change and agriculture, and what did exist—including Rosenzwieg’s own work—suggested that rising levels of carbon dioxide could have a “fertilizing” effect on some plants. That finding would ultimately become an enduring point in the effort to undermine climate science broadly—a “positive” seized upon by politicians and industry lobbyists as they sought to minimize the climate emergency.

But as the disciplines of crop and climate modeling evolved, they began to point more conclusively toward much more worrisome possible outcomes. If crops failed, especially in two or three major breadbasket regions at the same time, as some models began to suggest, millions of people could starve.   

And while famine and malnutrition are complicated problems, in the decades since these models began to examine the projected impact of global warming on food production, it’s become increasingly clear that climate change is a “threat multiplier,” making hunger emergencies worse. In some cases it could be the primary cause. 

Nearly 1 billion people went hungry or were malnourished last year and that number is projected to rise this year. 

Prompted in part by Rosenzweig’s work, a growing cadre of researchers started looking at combinations of other variables—including rain, soil quality, fertilizers, pests, carbon dioxide levels, crop varieties. The data improved. The models got more sophisticated. And, eventually, these scientists began to collaborate. 

In 2008, at a conference in Florida on water use in farming, Rosenzweig began to round up fellow scientists for what would eventually become the world’s biggest and most ambitious joint modeling effort to understand how climate change jeopardizes the agricultural systems that humans depend on for survival. 

About four years later, AgMIP researchers produced their first major paper.

The research said that the models “agreed” that the detrimental effects from climate change—mostly in developing countries around the planet’s midsection where more extreme weather events could batter crops—would be worse than previous research had suggested. It also emphasized some of the results were highly uncertain.

But now, after six more years of work, AgMIP researchers have bolstered those findings. Their latest major paper, which rests on improved models and updated climate data, projects a more alarming picture—one that will appear even sooner.

“More crops are predicted to react negatively,” said Jonas Jägermeyr, the lead author of the paper, which was published late last year in Nature Food.

Jägermeyr, a crop modeler and climate scientist, also at GISS, noted that the projected yields of corn dropped by more than 20 percent globally compared to current production levels. “That’s a completely new realm,” he said. “Across the world and in many bread basket regions, this is going to occur in the next couple years. The main message here is: This is right around the corner.”

Noting that the AgMIP modelers looked at roughly 8,000 simulations, Hasegawa said, “that gives us a better confidence.”

But even though researchers are increasingly confident that crop yields will falter, they say there’s a lot of work to be done in the modeling discipline. The world’s population will hit 10 billion people in 2050 when hotter temperatures and increased flooding will make feeding them more challenging. Knowing when and where the declines will happen—getting a full view of the risks—will be critical to preventing famine and malnutrition. 

“Modeling is essentially a way of creating transparency. In essence it gives us a view of something that we wouldn’t be able to see and couldn’t quantify without models,” said Molly Jahn, a plant geneticist, then a deputy secretary at the USDA, now at the Defense Advanced Research Project Agency (DARPA). “These models are not necessarily the right kind of models to do risk modeling.”

Food insecurity is an incredibly complex problem, not just the result of drops in the yields of major crops, but of politics, governance and economics. Climate change makes it only more complicated and urgent. The current models don’t account for all these factors yet.

Joshua Elliott, a program manager at DARPA who specializes in complex models, is one of dozens of researchers working on a new crop modeling system that goes beyond crop yield projections and weighs other factors, including political conflict and population flows.

“Our goal is to be able to improve the models,” Elliott said. “There’s just a massive amount of uncertainty. These are incredibly complex problems.”

In January, the United Nations said that last year 283 million people in 80 countries went hungry or were at high risk of going hungry—a record number—and more than 800 million were malnourished. Humanitarian aid groups have warned that the number of hunger emergencies in 2022 will very likely rise. 

The models can’t yet say where or how much.

Getting Serious About Crop Research and Modeling

If there’s a point at which the relatively esoteric science of crop modeling left the confines of its discipline, it was in the early 1970s, decades before AgMIP, when the Soviet Union made a huge deal to buy billions of dollars of U.S. wheat at prices that were cheap because of government subsidies.

U.S. negotiators at the time hadn’t realized that the Soviet Union’s wheat crops had failed and the deal took them by surprise, causing wheat shortages and a global price spike. 

After the “Great Grain Robbery,” as it was dubbed, the U.S. government started getting more serious about crop research and modeling in particular.  

Up to that point, most of the projections were made on statistical or mathematical models that looked at historical yields. But after the Russian grain purchase, the government developed models based on remote satellite sensing that could make strategic forecasts about crop yields. 

Over the next two decades, interest in crop modeling grew.  

“The heyday was in the ‘80s and ‘90s,” said David Fleisher, an agricultural engineer with the USDA who helps develop crop models. “There was tremendous development.”

Jerry Hatfield, a longtime USDA researcher and original AgMIP co-founder, remembers a moment in 1990 when the first global report on climate change was published by the IPCC.

“The IPCC originally came out and focused on rising CO2 levels and crop productivity and made a statement that all crops love CO2 so there won’t be a problem,” Hatfield said. “A lot of us sat around thinking: Let’s look at this system a little more holistically.”

Agricultural modeling needed a global approach—like the one the IPCC was taking for climate change writ large, the group of researchers concluded. In order to understand how climate change could shift or reduce the planet’s food supply, they needed to compare all the various models out there and, ultimately, improve them to get a clearer picture of the future.

“The results were too helter-skelter,” Rosenzweig said. “Different groups and scientists were saying, ‘We’re doing this scenario and we’re doing this baseline and we’re doing these projections and this model and that model.’ The IPCC was having a very hard time assessing the results of all those findings.”

The idea behind AgMIP was to put all the models into a harmonized “ensemble” and then feed them the same inputs (or data points) and parameters. 

“We found there wasn’t any single model that could help us predict what was happening in terms of productivity,” Hatfield said. “But if you took an ensemble of models—about 10 at a time—and you take the averages, they start to tell you something.”

Like the IPCC climate models, these crop models “talk” to each other.

“AgMIP was conceived to do for agriculture modeling” what these climate models did, said Sonali McDermid, a professor of environmental studies at New York University and an AgMIP researcher “The big IPCC reports that come out every four years—the science in those are informed by the [climate modeling] project that brings together all the world’s climate models, developed independently, and compares them.” 

AgMIP layers in these climate models, using their projections, to do roughly the same thing for agriculture. And in the study published last November, the AgMIP researchers found, with greater certainty, that most major crops would see reduced yields, though wheat yields could improve in northern latitudes in the short term. In some regions, the yield declines could happen more frequently within a decade, largely because increased heat will damage harvests.

“Once you execute all these models, you get a prediction, and this prediction is alarming,” said Bruno Basso, a professor of earth and environmental sciences at Michigan State University who specializes in crop and modeling research. “The threat is immense.”

But the AgMIP research, at least so far, doesn’t tell the whole story. It doesn’t yet account for steps farmers could take to adapt to changing climates, nor does it factor in economic incentives that could help push farmers to change their farming practices. (That research is forthcoming, the researchers note.)

“The thing about AgMIP that was transformative is, we were looking at models in the same way that a meteorologist would look at the path of a hurricane. You have a line,” said Lew Ziska, a former USDA plant physiologist, now a professor at Columbia University. “But when you put these models together, you get a much better forecast. That’s exactly what AgMIP does with respect to climate and food. That’s the good side of the coin.”

But, Ziska added, “the areas that need further elucidation are: What’s going to happen to food nutrition, what’s going to happen in terms of contamination of food, how might climate change affect pathogens. We have very strong evidence that climate change is going to adversely affect pesticides. It’s a good first step, but it isn’t a full description of all the challenges that need to be met.”

Some critics have also suggested that smaller-scale, statistical models—those based on historical crop yield, rather than projections made via simulations and supercomputing—are more useful because they produce results faster and are cheaper.  Others say that the type of global models used by AgMIP don’t fully capture the impact of climate change on wheat and rice.

Even the AgMIP research found that yields of rice and soybeans drop in some regions, but that the models don’t “agree” on the overall global impact.

Rosenzweig is aware of the limitations. “The ‘I’ is for improvement,” she jokes, referring to the AgMIP acronym.

There are important next-steps ahead. “What we really need at this point is to link the people out in the field, who know what’s going on in their region and the location realities, with the somewhat disconnected global climate communities and modeling community,” Jägermeyr, lead author of AgMIP’s latest paper, said. 

Other researchers agree. 

“It’s no good just running models and publishing results, or even communicating results to policy/society/industry,” Andrew Challinor, a professor at the University of Leeds and crop modeler, wrote in an email. “The stakeholders need to be involved right from the start.”

One major worry in the research community is about climate-induced “food shocks”—a sudden loss of a harvest that brings on a food shortage—that are more difficult to predict than the more gradual decline in crop yields that AgMIP has so far focused on.  

“In addition to the challenge of producing enough food on a global scale in 2050, we’re also going to be looking at a climate where we have much more year-to-year variability and we’re going to face a lot more agricultural production shocks in a lot of countries,”  said Chris Funk, director of the Climate Hazards Center at the University of California, Santa Barbara. “We used to have one crisis a year. Now we’re having three or four serious crises at the same time.”